Finalist: The New York Times

The promise was a war waged by all-seeing drones and precision bombs. The documents show flawed intelligence, faulty targeting, years of civilian deaths — and scant accountability.

By Azmat Khan

Shortly before 3 a.m. on July 19, 2016, American Special Operations forces bombed what they believed were three ISIS “staging areas” on the outskirts of Tokhar, a riverside hamlet in northern Syria. They reported 85 fighters killed. In fact, they hit houses far from the front line, where farmers, their families and other local people sought nighttime sanctuary from bombing and gunfire. More than 120 villagers were killed.

In early 2017 in Iraq, an American war plane struck a dark-colored vehicle, believed to be a car bomb, stopped at an intersection in the Wadi Hajar neighborhood of West Mosul. Actually, the car had been bearing not a bomb but a man named Majid Mahmoud Ahmed, his wife and their two children, who were fleeing the fighting nearby. They and three other civilians were killed.

In November 2015, after observing a man dragging an “unknown heavy object” into an ISIS “defensive fighting position,” American forces struck a building in Ramadi, Iraq. A military review found that the object was actually “a person of small stature” — a child — who died in the strike.

None of these deadly failures resulted in a finding of wrongdoing.

These cases are drawn from a hidden Pentagon archive of the American air war in the Middle East since 2014.

The trove of documents — the military’s own confidential assessments of more than 1,300 reports of civilian casualties, obtained by The New York Times — lays bare how the air war has been marked by deeply flawed intelligence, rushed and often imprecise targeting, and the deaths of thousands of civilians, many of them children, a sharp contrast to the American government’s image of war waged by all-seeing drones and precision bombs.

The documents show, too, that despite the Pentagon’s highly codified system for examining civilian casualties, pledges of transparency and accountability have given way to opacity and impunity. In only a handful of cases were the assessments made public. Not a single record provided includes a finding of wrongdoing or disciplinary action. Fewer than a dozen condolence payments were made, even though many survivors were left with disabilities requiring expensive medical care. Documented efforts to identify root causes or lessons learned are rare.

The air campaign represents a fundamental transformation of warfare that took shape in the final years of the Obama administration, amid the deepening unpopularity of the forever wars that had claimed more than 6,000 American service members. The United States traded many of its boots on the ground for an arsenal of aircraft directed by controllers sitting at computers, often thousands of miles away. President Barack Obama called it “the most precise air campaign in history.”

This was the promise: America’s “extraordinary technology” would allow the military to kill the right people while taking the greatest possible care not to harm the wrong ones.

The ISIS caliphate ultimately crumbled under the weight of American bombing. For years, American air power was crucial to the beleaguered Afghan government’s survival. And as U.S. combat deaths dwindled, the faraway wars, and their civilian tolls, receded from most Americans’ sights and minds.

On occasion, stunning revelations have pierced the silence. A Times investigation found that a Kabul drone strike in August, which American officials said had destroyed a vehicle laden with bombs, had instead killed 10 members of one Afghan family. The Times recently reported that dozens of civilians had been killed in a 2019 bombing in Syria that the military had hidden from public view. That strike was ordered by a top-secret strike cell called Talon Anvil that, according to people who worked with it, frequently sidestepped procedures meant to protect civilians. Talon Anvil executed a significant portion of the air war against ISIS in Syria.

The Pentagon regularly publishes bare-bones summaries of civilian casualty incidents, and it recently ordered a new, high-level investigation of the 2019 Syria airstrike. But in the rare cases where failings are publicly acknowledged, they tend to be characterized as unfortunate, unavoidable and uncommon.

In response to questions from The Times, Capt. Bill Urban, the spokesman for the U.S. Central Command, said that “even with the best technology in the world, mistakes do happen, whether based on incomplete information or misinterpretation of the information available. And we try to learn from those mistakes.” He added: “We work diligently to avoid such harm. We investigate each credible instance. And we regret each loss of innocent life.”

He described minimizing the risk of harm to civilians as “a strategic necessity as well as a legal and moral imperative,” driven by the way these casualties are used “to feed the ideological hatred espoused by our enemies in the post 9/11 conflicts and supercharge the recruiting of the next generation of violent extremists.”

Yet what the hidden documents show is that civilians have become the regular collateral casualties of a way of war gone badly wrong.

To understand how this happened, The Times did what military officials admit they have not done: analyzed the casualty assessments in aggregate to discern patterns of failed intelligence, decision-making and execution. It also visited more than 100 casualty sites and interviewed scores of surviving residents and current and former American officials. In the coming days, the second part of this series will trace those journeys through the war zones of Iraq and Syria.

Taken together, the reporting offers the most sweeping, and also the most granular, portrait of how the air war was prosecuted and investigated — and of its civilian toll.

There is no way to determine that full toll, but one thing is certain: It is far higher than the Pentagon has acknowledged. According to the military’s count, 1,417 civilians have died in airstrikes in the campaign against ISIS in Iraq and Syria; since 2018 in Afghanistan, U.S. air operations have killed at least 188 civilians. But The Times’s analysis of the documents found that many allegations of civilian casualties had been summarily discounted, with scant evaluation. And the on-the-ground reporting — involving a sampling of cases dismissed, cases deemed “credible” and, in Afghanistan, cases not included in the trove of Pentagon documents — found hundreds of deaths uncounted.

The war of precision did not promise that civilians would not die. But before a strike is approved, the military must undertake elaborate protocols to estimate and avoid civilian harm; any expected civilian casualties must be proportional to the military advantage gained. And America’s precision bombs are indeed precise: They hit their targets with near-unerring accuracy.

The documents, along with The Times’s ground reporting, illustrate the many, often disastrous ways the military’s predictions of the peril to civilians turn out to be wrong. Their lessons rarely learned, these breakdowns of intelligence and surveillance occur again and again.

Repeatedly the documents point to the psychological phenomenon of “confirmation bias” — the tendency to search for and interpret information in a way that confirms a pre-existing belief. People streaming toward a fresh bombing site were assumed to be ISIS fighters, not civilian rescuers.

Men on motorcycles moving “in formation,” displaying the “signature” of an imminent attack, were just men on motorcycles.

Often, the danger to civilians is lost in the cultural gulf separating American soldiers and the local populace. “No civilian presence” was detected when, in fact, families were sleeping through the days of the Ramadan fast, sheltering inside against the midsummer swelter or gathering in a single house for protection when the fighting intensified.

In many cases, civilians were visible in surveillance footage, but their presence was either not observed by analysts or was not noted in the communications before a strike. In chat logs accompanying some assessments, soldiers can sound as if they are playing video games, in one case expressing glee over getting to fire in an area ostensibly “poppin” with ISIS fighters — without spotting the children in their midst.

The military spokesman, Captain Urban, pointed out that, “In many combat situations, where targeteers face credible threat streams and do not have the luxury of time, the fog of war can lead to decisions that tragically result in civilian harm.”

Indeed, the Pentagon records detail how in Mosul in 2016, three civilians were killed when a bomb aimed at one car instead struck three — in part because the military official approving the strike had decided to save more-precise weapons for other, imminent strikes. Yet The Times’s analysis of the documents and ground reporting showed that civilians were frequently killed in airstrikes planned well in advance.

Military officials often speak of their “over the horizon” long-range surveillance capabilities. But the documents repeatedly identify deficiencies in the quality and quantity of the video footage guiding intelligence.

Sometimes, only seconds’ worth of footage was taken before a strike, hardly enough to assess civilians’ presence. Often video shot from the air does not show people inside buildings, people under foliage, people under the aluminum or tarpaulin covers known as “quamaria” that shield cars and market stalls from the sun.

In more than half of the cases deemed credible by the military, one or two civilians were killed entering the target area after a weapon was fired. Officials often describe these as awful but inescapable accidents. But while many might have been averted through additional precautions — widening the surveillance camera’s field of view or deploying additional drones — the phenomenon continued unabated, amid the intense pace of battle and a shortage of surveillance aircraft.

And sometimes, for reasons redacted in the documents, the weapons simply miss. In April 2016 the military reported that it had killed a notorious Australian ISIS recruiter, Neil Prakash, in a strike on a house in East Mosul. Months later, very much alive, he was arrested crossing from Syria into Turkey. Four civilians died in the strike, according to the Pentagon.

Yet despite this unrelenting toll, the military’s system for examining civilian casualties rarely functions as a tool to teach or assess blame.

Not only do the records contain no findings of wrongdoing or disciplinary action, but in only one instance is there a “possible violation” of the rules of engagement. That stemmed from a breach in the procedure for identifying a target. Full investigations were recommended in fewer than 12 percent of the credible cases.

In many cases, the command that approved a strike was responsible for examining it, too. And those examinations were often based on incorrect or incomplete evidence. Military officials interviewed survivors or witnesses in only two cases. Civilian-casualty reports were regularly dismissed because video showed no bodies in the rubble, yet the footage was often too brief to make a true determination.

In his response to The Times, Captain Urban said, “An honest mistake, on a strike taken with the best available information and in keeping with mission requirements that results in civilian casualties, is not, in and of itself, a cause for disciplinary actions as set forth in the law of armed conflict.”

American officials had an opportunity to mine the documents for root causes and patterns of error in 2018, when the Joint Chiefs of Staff and the National Defense University undertook a study of civilian deaths. But one of the researchers who sought to analyze the documents in aggregate told The Times that almost all of his findings had been cut from the report. Another high-level study of the air campaign has never been made public.

In the end, what emerges from the more than 5,400 pages of records is an institutional acceptance of an inevitable collateral toll. In the logic of the military, a strike, however deadly to civilians, is acceptable as long as it has been properly decided and approved — the proportionality of military gain to civilian danger weighed — in accordance with the chain of command.

Lawrence Lewis, the former Pentagon and State Department adviser whose analysis for the 2018 study was quashed, said in an interview that the military’s technological prowess, and the highly bureaucratized system for assessing how it is employed, may actually serve an unspoken purpose: to create greater legal and moral space for greater risk.

“Now we can take strikes in city streets, because we have Hellfire missiles, and we have fancy things with blades,” he said. “We develop all these capabilities, but we don’t use them to buy down risk for civilians. We just use them so we can make attacks that maybe we couldn’t do before.”

The Promise of Precision

The new way of war came to fruition in the wake of the 2009 surge of American troops into Afghanistan, which brought some stability but never turned the war around.

By the end of 2014, with NATO’s mission also ending, President Obama declared America’s ground war essentially done. Henceforth, the United States would primarily provide air support and advice for Afghan forces battling the Taliban.

At roughly the same time, as Islamic State fighters swept through Mosul and massacred thousands of Yazidi Kurds at Mount Sinjar, Mr. Obama authorized a campaign of airstrikes against ISIS targets and in support of allied forces in Iraq and Syria.

The weaponry was hardly untested. This high-tech arsenal, increasingly sophisticated, had been critical to success in the 1991 Persian Gulf war, in NATO’s 1999 campaign in the Balkans, and more recently in Yemen and Somalia. By the time of the wars in the Middle East, the MQ-9 Reaper drone, outfitted with laser-guided Hellfire missiles, had become the surveillance and attack vehicle of choice.

At an ever-quickening pace over the next five years, and as the administration of Mr. Obama gave way to that of Donald J. Trump, American forces would execute more than 50,000 airstrikes in Iraq, Syria and Afghanistan, in accordance with a rigorous approval process that prized being “discriminate,” “proportional” and in compliance with the law of armed conflict. Not only would this be the most precise air campaign ever; it would be the most transparent.

The only official accounting of that promise is the hidden Pentagon documents.

They were obtained through Freedom of Information requests beginning in March 2017 and lawsuits filed against the Defense Department and U.S. Central Command. To date, The Times has received 1,311 out of at least 2,866 reports — known as credibility assessments — examining airstrikes in Iraq and Syria between September 2014 and January 2018. Requests for records from Afghanistan are the subject of a new lawsuit.

Each report is the fruit of a review process that begins when a potential civilian-casualty incident is identified by the military or, more frequently, alleged by an outside source — a nongovernmental organization, a news outlet or social media.

Assessment experts classify allegations into two categories. A case is “credible” if it is deemed “more likely than not” that the airstrike caused civilian casualties. In the reports examined by The Times, 216 cases were deemed credible. “Noncredible” cases fail to meet that standard — often because there is no record of a strike at the place and time in question, or because the available evidence is considered insufficiently specific or simply weak.

Until now, fewer than 20 of these assessments dating to late 2014 have been made public.

To assess the military’s assessments, between late 2016 and this past June, The Times visited the sites of 60 incidents deemed credible in Iraq and Syria, as well as three dozen others deemed noncredible or not yet assessed. (It also visited dozens of strike sites in Afghanistan.) In 35 credible cases, it was possible to locate the precise impact area and find survivors and witnesses on the ground. Then the reporting included touring wreckage; collecting photo and video evidence; and verifying casualties through death certificates, government IDs and hospital records.

Frequently the reporting closely matched basic information from the documents. But the detailed accounts that ultimately emerged from the rubbled ground were often in stark contrast to what had been assessed from the air.

‘Play Time?’

this area is poppin

It was Jan. 13, 2017, and the battle for East Mosul would soon reach the neighborhood of al-Faisaliya. Iraqi forces were 120 meters away; farther back, an American ground team was helping coordinate air support.

In Erbil and in Doha, Qatar, a ground controller and aircrew members typed out messages, helping fulfill the array of combat directives and rules of the strike process:

adm in kp 9 has his rifle leaning against wall

An adult male leaned against a rooftop wall, his rifle beside him, then was seen firing south before two men joined him.

play time?

The ground controller asked how much longer the crew had in the target area. The response was redacted.

A man was seen running into a building, then out.

bldg slant redacted

The “slant” — the number of men, women and children observed — was typed into the chat. (Four men, one woman and three children in a building would be “bldg slant 4/1/3.”) This slant is redacted.

The coordinates were entered for what was now assessed as a building used by ISIS.

cleared hot

Clearance to attack was granted, and the weapon — the exact kind is redacted — fired.

splash

Five seconds to impact.

Two “squirters” — people fleeing a bomb site — were observed: one running from the building, the other heading back inside. The drone followed the men, firing on one but overshooting. It fired again, then turned to four others.

The action continued — a series of attacks on men darting through the area, until the drone returned to the building and struck again.

bldg was completely dropped

Toward the end, men were observed getting into a van.

looks like children as well

The war against ISIS heralded the dawn of “strike cells” — remote operations centers from which most airstrikes were directed and controlled. These war rooms synergized the myriad players — pilots, sensor operators, intelligence experts, ground forces, weaponeering specialists, civilian-casualty-mitigation analysts, lawyers, even weather officers. Strike cells boasted at times that, with their video feeds and surveillance aircraft, they could understand what was happening on the battlefield as well as if they were there themselves.

As the war intensified and ground commanders won greater authority to call in strikes, the cells expanded, with a small number of Americans embedded with allies on the battlefield. The cells were seen as so successful that they made their way to Afghanistan, too. And as the Trump administration sought to pressure the Taliban into a deal, decision-making authority for airstrikes was often pushed further down the chain of command.

The cells conducted “dynamic strikes” — identified and executed within minutes or hours in the flow of war, accounting for an overwhelming majority of the air campaign. “Deliberate” strikes, which were preplanned — extensively vetted, often filmed over weeks or months and analyzed by several working groups — decreased over time.

In both scenarios, the targeting process essentially boiled down to two questions: Could the presumed enemy target be positively identified? And would any harm to civilians be proportional, in line with the law of armed conflict — or would it exceed the “expected military advantage gained”?

For positive identification, the officer designated with strike approval needed “reasonable certainty” that the target performed a function for the adversary. That could be relatively straightforward, as when the target was a fighter firing directly on friendly forces. But a more ambiguous target, like a suspected ISIS headquarters, might require further surveillance.

To determine proportionality, analysts evaluated whether the target was used exclusively by the enemy or might also be used by civilians, then assessed civilians’ “pattern of life.” Ultimately, they would calculate how many civilians were likely to be killed or wounded.

For deliberate strikes, this generally entailed an exhaustive “collateral damage estimate,” a computer calculation of the expected civilian casualty count, based on a mix of factors: the pattern of life, the population density, the specific weapon being used, the kind of structure being targeted — a concrete building, an aluminum shed, a mud hut. The officer approving the strike would weigh that estimate with other factors, such as the potential for secondary blasts from explosive materials nearby.

For dynamic strikes, the process could be vastly compressed. Especially if there was a threat to friendly forces or some other urgency, strike cells were more likely to rely on an impromptu assessment of a video feed.

Either way, based on that calculation, the military was required to take “feasible precautions” to mitigate civilian harm. The greater the likelihood of someone being in the wrong place at the wrong time, the more precautions taken — say, by deploying more-precise weaponry to limit the blast radius or by attacking when the fewest civilians were predicted to be present.

The military does not provide a precise definition of what is proportional. Essentially, the expected civilian toll was proportional if the officer making that determination reasonably believed it to be so, and if it did not exceed a “noncombatant cutoff value.” Otherwise, officials say, the target would be discarded.

The final official step was a legal review. But efforts to protect civilians could continue until moments before a weapon was fired. From the cockpit, pilots could select how a weapon detonated — upon impact or with a delayed fuse. Or they could call an “abort,” if, for example, a civilian was spotted walking into the target area.

Under the right circumstances, this process could result in a strike so precise that it would destroy the section of a house filled with enemy fighters and leave the rest of the building intact.

As Iraqi forces approached Qusay Saad’s home in East Mosul, ISIS forced his family to move to an area still under its control. They found refuge in his brother’s abandoned house in al-Faisaliya.

Through a night of gunfire and explosions, Mr. Saad and his wife, Zuhour, comforted their three children and prayed that Iraqi forces would reach them. Then ISIS ordered them to move again, into an abandoned school next door with two other families. That was the building observed in the chat on Jan. 13, 2017.

The first airstrike hit as the Saad family sat down to breakfast. Mr. Saad recalls concrete blocks pressing down on his head, and his wife screaming. A man from one of the other families lifted away the blocks, and he quickly wrested his 14-month-old daughter, Aisha, from the rubble and handed her to his wife.

The second strike came just as he turned to free his 7-year-old son, Muhammad.

“The strike was unbelievable,” he said. “An entire three-story house was just crushed.”

Three members of another family escaped. Mr. Saad could not find his wife, their 4-year-old son, Abdulrahman, or Aisha. But Muhammad was alive, his thigh split open. Bleeding from the head, Mr. Saad picked up the boy and fled.

It would be two months before he could recover the bodies. The Iraqi government offered no help. So the family paid to excavate the site. Mr. Saad watched as his wife and two youngest children were lifted out. Aisha’s head was missing, but her little body was in her mother’s arms.

They were buried not far from their home, which Mr. Saad has kept as it was when they all lived there. Sometimes, his brother said, he spends whole nights at the graveyard.

Last month, The Times told him of the findings of the military’s assessment. It offers this account:

The target was a building assessed as harboring four ISIS fighters. A review of the imagery revealed that after the first strike, which because of a “weapon malfunction” only partly collapsed the building, four adults and four children could be seen moving in its center. The building was hit again and fully collapsed. Later, three people emerged. The strike team did not report any civilians in the vicinity, and because of the drone’s angle, a view of the eight people in the building after the first strike “was obscured.”

The allegation was deemed credible, with eight civilians killed, but no further investigation was ordered. Eight “enemies” were also killed, the document said.

How Deadly Failures Happen

Last May, the Pentagon’s inspector general completed a classified report evaluating the policies for ensuring that “only valid military targets are struck,” and that “damage to property and loss of civilian life is mitigated to the maximum extent possible.”

A redacted version, echoing similar studies by other agencies in recent years, declares the targeting process to be sound.

The Pentagon’s own assessments tell a far richer story.

The documents often do not articulate precise causes, and in many cases, several factors coalesced into a deadly failure. But The Times’s analysis of the 216 cases deemed credible, together with its reporting on the ground, reveals several distinct patterns of failure.

Misidentifying Civilians

Positive identification of the enemy is one of the pillars of the targeting process, yet ordinary citizens were routinely mistaken for combatants.

In a dissenting footnote to the 2018 Joint Chiefs’ study, Mr. Lewis and a colleague cited research showing that misidentification was one of the two leading causes of civilian casualties in American military operations. With few troops on the ground, they wrote, “it is reasonable to expect a systematic undercounting of misidentifications in U.S. military reports.”

Indeed, according to the Pentagon records, misidentification was involved in only 4 percent of cases. At the casualty sites visited by The Times, misidentification was a major factor in 17 percent of incidents, but accounted for nearly a third of civilian deaths and injuries.

At times, the error involved quicksilver intelligence of an imminent threat. In The Times’s ground sample, though, misidentification occurred just as frequently in strikes planned far in advance — as in a January 2017 strike on an ISIS “foreign fighter headquarters” in East Mosul that killed 16 people in what turned out to be three civilian homes. Three ISIS buildings down the street were untouched.

Yet in case after case, the misidentification appears to be less a matter of confusion than of confirmation bias.

That was what happened on Nov. 20, 2016, after a Special Operations task force received a report of an ISIS explosives factory in a Syrian village north of Raqqa. In a walled compound, operators spotted “white bags,” assessed to be ammonium nitrate. Two trucks with a dozen men departed, stopped at various ISIS checkpoints, drove to a building “associated with previous ISIS activity,” then returned to the compound. The first strike targeted one truck, which caused “secondary explosions.” On the evidence of those blasts and the “white bags,” operators received approval to strike three buildings. After impact, two “squirters” fled the westernmost building. That building and another were struck again.

The findings of the military’s review, begun after online reports that a strike in the same area had killed nine civilians and injured more than a dozen, contradicted nearly all of the original intelligence.

“I remember there was a big explosion, and I fainted,” recalled Abdul Hakeem Abdullah Hamash Al Aqidi, an elderly man who had been standing by his door at the intersection. He lost an eye and had to have a plate implanted in his injured left leg. His son’s left leg had to be amputated.

In all, seven local people — including the four members of the Ahmed family — were killed. Mr. Mohamed, who had waved to Mr. Ahmed, cannot banish from his mind the image of his friend’s wife, Hiba Bashir, burned into the seat, still holding her infant son in her lap.

The military spokesman, Captain Urban, acknowledged that “confirmation bias is a real concern,” citing the Kabul airstrike in August that killed the 10 members of a family. “There is more work to do on this,” he said.

Failing to Detect Civilians

If the military often mistook civilians for enemy fighters, frequently it simply failed to see or understand that they were there. That was a factor in a fifth of the cases in the Pentagon documents, and a slightly smaller fraction of the casualties. However, it accounted for 37 percent of credible cases, and nearly three-fourths of the total civilian deaths and injuries at the sites visited by The Times.

Captain Urban said the targeting process had been vastly complicated by enemies who “plan, resource and base themselves in and among local populace.”

“They do not present themselves in large formations,” he added, “do not fight coalition forces with conventional tactics, and use geography and terrain in ways not conducive in every way to easy targeting solutions. Moreover, they often and deliberately use civilians as human shields, and they do not subscribe to anything remotely like the law of armed conflict to which we subscribe.”

Even so, the documents show that frequently, instead of extended surveillance, analysts relied on brief “collateral scans” — as little as 11 seconds long — in determining that civilians were not in the area. The footage was often limited by shortages of surveillance drones, particularly during the battles to retake Mosul and Raqqa.

In a number of cases, targets that had been placed on “no-strike lists” because attacking them would violate laws of war — a school, a bakery, a civilian hospital — were removed after the military mistakenly judged that they were now used exclusively by the enemy.

In Mosul in February 2017, a hospital was taken off the list after the military concluded that civilians had left the area, and that the building was being used only as an ISIS headquarters and propaganda center. The week before the strike, according to the report, analysts had examined still images of children “interacting” with the hospital but had determined that striking at night would “alleviate collateral concerns.” Four civilians were killed and six injured.

For the military’s analysts, studying the “pattern of life” is a crucial step in predicting collateral damage. But to examine the documents and interview local people is to understand how often unseen civilians might have been seen, or their presence at least suspected, had the military had a more intimate knowledge of the war-torn fabric of everyday life.

In some documents, as evidence of no civilian presence, military officials state that people would leave their homes at the sound of approaching aircraft. The reality is starkly different: Neighbors would huddle together, seeking communal sanctuary in a house or group of houses, invisible to surveillance drones.

Many of the deadliest airstrikes happened this way. Among them was the strike at the Syrian hamlet of Tokhar.

In July 2016, a Special Operations task force identified a large group of ISIS fighters two kilometers from where U.S.-backed forces were fighting ISIS. They observed the fighters traveling in pickups known as “bongo trucks” to three “staging areas” where no civilians were present. The fighters, they concluded, were assembling for a counterattack. Shortly before 3 a.m., they bombed the three staging sites and five vehicles, confident of killing 85 ISIS fighters.

Almost immediately, reports of a vast civilian death toll surfaced online. The task force conducted a full investigation and determined that between seven and 24 civilians “intermixed with the fighters” might have been killed.

The Times visited Tokhar in December 2018. Surviving villagers gave this account:

That night, as they had every night for a month, some 200 villagers had trekked to the outer edge of the hamlet and taken shelter in four homes at the farthest remove from the quickening battle.

There was no evidence, they said, that ISIS had been near any of the four houses. In fact, residents said drones had been flying overhead for weeks, giving them solace that coalition forces knew they were there.

The Times documented the names of civilians killed in each of the four houses, corroborating details with open-source information, local journalists and others on the ground, and determined that more than 120 people died. There were few young men left to pull bodies from the rubble. It took nearly two weeks, and still some were never found. If the full death toll were acknowledged, Tokhar would be the largest civilian casualty incident the United States has admitted to in the air war against ISIS.

Saif Saleh, 8 years old at the time, awoke that early morning to the collapsing walls, his arm trapped under debris. His parents used up every favor to collect $6,000 for surgery to graft skin from his leg.

Asked what he would like to tell the American military, Saif’s father said, “We want to say that you should be sure the area is empty or that there are no civilians before you bomb.”

The military investigation found that there was no evidence of negligence or wrongdoing; that the “policies, procedures and practices” were “sufficient for continued operations”; and that “no further action” was necessary. No condolence payments were authorized.

Overlooking Flawed Intelligence

Often, civilians were killed in strikes executed in the face of incomplete, outdated or ambiguous intelligence. Several such factors came together in a strike that killed at least 10 civilians in Tabqa, Syria, in March 2017.

As American-backed forces prepared to recapture the city, west of Raqqa, military officials approved strikes on a group of ISIS targets: two headquarters, a police station and a weapons factory. Each strike went as planned, according to initial assessments. Then came reports of civilian casualties.

The military review found that the intelligence for both headquarters was based on single reports from months before. (The targets had been identified earlier, but for strategic advantage, commanders had decided to wait until Syrian Democratic Forces were pushing into Tabqa.) The intelligence package on the first building warned that there was “insufficient” evidence to corroborate the judgment, relied on to remove the building from a restricted-targeting list, that it was used solely by ISIS; the report said simply that an ISIS emir had frequented the site.

Similarly, the review found that the intelligence did not support the view that the second headquarters was used exclusively by ISIS. What’s more, even though both headquarters were in densely populated areas with residential structures nearby, there was insufficient footage to assess the presence of civilians — one minute of video of the first target and less than two of the second.

The review also raised serious questions about the quality of intelligence for the two other targets.

Flawed Video

Sometimes, the problem was less the quantity of video than the quality.

Analysts at the military’s Combined Air Operations Center in Qatar saw this clearly when they reviewed 17 minutes of grainy footage that preceded a Nov. 13, 2015, strike on an ISIS “defensive fighting position” in Ramadi. Using the center’s 62-inch high-definition TV, they concluded that what had been identified as an “unknown heavy object” being dragged into a building was actually “a person of small stature,” “consistent with how a child would appear standing next to an adult.”

Often the overhead surveillance camera missed people simply sitting or standing under something, doing the most quotidian things.

June 15, 2016: An ISIS fighter on a motorcycle turned onto a secondary road near Mosul University. It was Ramadan; the shops and stalls were teeming with people. Among the five civilians also killed and four wounded in the strike:

Abdul Wahab Adnan Qassim, killed by shrapnel, had been standing in the tree-filled courtyard of his house.

Zanoun Ezzedine Mahmoud, killed by the blast, had been standing at a fruit stand covered by a blue tarp blocking the sun. The stand’s owner, Ilyas Ali Abd Ali, lost his right leg.

A father and daughter, killed by glass and shrapnel, had been sitting in a car nearby.

Nashwan Abdul Majeed Abdul Hakeem Al Radwani, killed by shrapnel, had been standing under the awning of the popular Hammurabi Ice Cream Shop.

Walking Into Danger

More than half of the cases the military deemed credible involved someone entering the target frame in the moments between a weapon’s firing and impact, as in a March 2017 strike in Mosul when shrapnel killed a man pushing a cart down a road near an ISIS mortar tube.

These deaths, which account for 10 percent of acknowledged civilian casualties, are often framed as unavoidable accidents. In the Mosul strike that killed the man with the cart, operators had already twice aborted weapons releases because civilians had entered the frame — demonstrating concerted efforts to avert danger. Yet the systematic nature of the problem suggests the military could be doing more.

Indeed, the review of a February 2017 strike on a “high value individual” at a funeral in Mosul that injured two civilians includes some recommendations. While noting that the civilians’ presence “could not be predicted to reasonable certainty,” it adds that an additional surveillance aircraft could have provided a more encompassing view. (Because of the target’s importance, two aircraft were used to zoom in, rather than out, on the wider scene.) Yet again surveillance drones were in short supply.

Secondary Explosions

In the late spring of 2015, as ISIS continued to prove resilient in carrying out attacks and retaining territory, American targeteers and weapons specialists prepared a nighttime airstrike on a car-bomb factory in the industrial district of Hawija, north of Baghdad. Occupied apartment houses ringed the area. But the nearest “collateral concern” was assessed to be a “shed.”

Not long before, dozens of displaced families, unable to afford rent, had also begun squatting in the abandoned houses scattered through the industrial zone. Among them were Khadijah Yaseen and her family, who had fled the fighting in their hometown, Yathrib.

The night of June 2 was particularly hot, so the family slept outside. They woke to screaming and the sound of the jets.

“There was fire everywhere,” Ms. Yaseen recalled when The Times met her at a displaced persons camp in October 2016. Most of those killed were from squatter families like hers. “You couldn’t count them. There were so many people that died.”

As many as 70, a military investigation found. Ms. Yaseen lost three grandchildren: 13-year-old Muhammad, 12-year-old Ahmed and a 3-year-old girl, Zahra.

Hawija is among the deadliest examples of the failure to predict the collateral consequences of striking weapons caches or other targets with the potential for secondary explosions. Such explosions often reached far beyond the expected blast radius; they accounted for nearly a third of all civilian casualties acknowledged by the military and half of all civilian deaths and injuries at the sites visited by The Times.

Although the American military planned the Hawija strike, the bombs were dropped by the air force of the Netherlands. There, the case became a cause célèbre after it emerged that the defense minister had worked to suppress the findings of the military investigation.

In the report of the investigation, targeteers and weapons experts describe the ultimately disastrous calculations taken to win approval for the strike. They worked and reworked the target, carefully calculating what kinds of munitions to use until their model concluded — despite the fact that they would be striking a car-bomb factory with apartment buildings nearby — that there would be no civilian deaths. (The Dutch military would only carry out strikes with an expected civilian-casualty rate of zero.)

The document describes a secondary explosion that produced a “visible shock wave” extending more than 750 feet from the target.

“That is massive, to be able to see a shock wave like that on a video,” said a former high-level official involved in the air campaign against ISIS, speaking on the condition of anonymity for fear of retaliation. The only comparable explosion he’d seen, he said, was the 2020 blast that devastated the port of Beirut.

Among the sites visited by The Times, at least half of the strikes with secondary explosions involved targets — like a power station or a factory for improvised explosive devices — that the military could have predicted would produce such blasts.

However, at other times it was unaware of both a weapons cache and a civilian presence. That was the case in the largest civilian casualty incident the military has admitted in the war, the March 17, 2017, airstrike on two ISIS snipers in the Mosul al-Jadida neighborhood that killed at least 103 civilians.

Failures of Accountability

On Jan. 6, 2017, Rafi Al Iraqi woke to the sound of a bomb close by. Another hit next door. Moments later, his own house was struck. He could hear his oldest son, Hamoody, screaming in the wreckage. “I just gave him to some people to take him to the hospital,” Mr. Al Iraqi recalled. “Then I went back in to find my other children.”

What happened next was captured on video taken by ISIS’ media agency, which often visited blast sites for propaganda.

Rescuers emerged holding limp bodies. Mr. Al Iraqi’s daughter, Nour, was alive. “I took her with my own hands to the hospital,” he recalled this past June, in his most recent interview with The Times. “But by then, she had died.” A nearby house for ISIS fighters was untouched.

Soon, via the ISIS video and news reports, word spread online that three families had been targeted in the Zerai neighborhood near Mosul’s Grand Mosque. In all, 16 civilians were killed, including three of Mr. Al Iraqi’s children and his mother-in-law. Hamoody’s leg was lacerated.

The military began a civilian-casualty assessment, which found that there had been a single strike in Zerai that day — on a house assessed to be used exclusively as an ISIS “foreign fighter headquarters” and “artillery staging location.” The strike had been preplanned, with no expected civilian casualties.

The post-strike footage showed no civilians killed or injured. The post-strike chat did not indicate the presence of civilians, though it did mention a wounded man — judged to be an ISIS fighter — being helped from the ruins.

The footage was 1 minute and 22 seconds long. The allegation was deemed noncredible. Officially, 16 people had not died that day in Zerai. (The Pentagon finally acknowledged the casualties in September 2020, after years of follow-up by The Times.)

Except for the rare instances of revelation and subsequent outcry, the Pentagon’s brief published reports on the minority of cases it finds credible are the only public acknowledgment of the air war’s civilian toll.

The Times’s reporting in Iraq, Syria and Afghanistan points to the broader truth.

In addition to the finding that many allegations of civilian casualties were erroneously dismissed, The Times discovered that even when civilian deaths were acknowledged, they were often significantly undercounted.

Roughly 37 percent of the allegations deemed credible stemmed from prior ground investigations by journalists or nongovernmental organizations; in those cases, the acknowledged death tolls roughly tracked outside reporting. But in the other cases, The Times’s own reporting found that the civilian toll was nearly double that acknowledged by the military. (That did not include ISIS fighters’ wives and children, whose information was difficult to verify.)

The documents identify children killed or injured in 27 percent of cases; in The Times’s ground reporting it was 62 percent. In 40 percent of the sites visited, survivors had been left with significant disabilities, which were not tracked by the military.

Beyond the casualty count, the structure and execution of the assessments do not encourage the regular examination of immediate lessons or deeper trends.

The records obtained by The Times, some significantly redacted, range from short first-impression reports to more formal credibility assessments. The reports generally contain a narrative drawn from the strike’s “target package” — including intelligence about the target, the civilian-casualty estimate, actions to mitigate civilian harm, video footage and chat logs tracking each step of the process.

Not only was there no record of disciplinary action, or full investigations in roughly 9 of every 10 cases, but only a quarter included any further review, recommendations or lessons learned. Even the architecture of the forms makes it difficult to analyze causes in aggregate; they do not have specific boxes for specific factors involved in a fatal error. There are a few places to record proximate causes or lessons learned, but those fields are mostly empty or redacted. Records are often incomplete, missing attachments or were only partially entered into shared databases.

In many cases, the unit that executed a strike also ended up investigating it; their assessments often included minimal information. For example, a Special Operations unit’s rationale for rejecting allegations that a December 2016 airstrike near Raqqa had killed as many as nine civilians consisted of a single paragraph stating that it had reviewed its strikes in the area and found “no evidence of possible civilian casualties.” There was no further information or detail from the footage.

The Times found that such omissions, as well as redactions and missing documents, were often associated with Talon Anvil, the Special Operations unit that carried out the recently revealed airstrike that killed dozens of civilians in Syria in 2019.

The Video Evidence

Of the 1,311 assessments from the Pentagon, in only one did investigators visit the site of a strike. In only two did they interview witnesses or survivors.

Captain Urban, the military spokesman, said that in hostile territory, investigators might be unable to visit a blast site and interview “personnel on the ground.”

Instead, often the resounding piece of evidence studied was video recorded in the wake of a strike. Yet just as poor or insufficient footage frequently contributed to deadly targeting failures, so did it hamstring efforts to examine them.

Often, the footage was only seconds or minutes long, in many cases too brief to see rescuers carrying survivors from a collapsed building. (Frequently, rescuers would wait before approaching a bombed area, for fear of being misidentified and provoking a second strike, known in the military as a “double tap.”) Often, images were obscured by the smoke of the blast.

In an interview — speaking anonymously because of a nondisclosure agreement — an analyst who captures strike imagery said superior officers would often “tell the cameras to look somewhere else” because “they knew if they’d just hit a bad target.”

And at times, there was simply no footage for review, which became the basis for rejecting the allegation. That was often because of “equipment error,” because no aircraft had “observed or recorded the strike,” or because the unit could not or would not find the footage or had not preserved it as required.

In a number of cases, compelling allegations were dismissed because the claim’s details did not precisely match the imagery.

For example, when Airwars — the leading source of civilian casualty allegations referred to the military — reported that a strike in East Mosul in April 2015 had killed dozens of civilian rescuers, the allegation was rejected because of “discrepancies in eyewitness accounts.” Despite accurately testifying that three bombs had struck an electric substation, a witness said the third had come a quarter-hour after the second and had not exploded; the document described that as “inconsistent” with the military’s imagery and strike report. (The allegation was later deemed credible after The Times visited the site and told the military that at least 18 civilians had been killed and more than a dozen wounded.)

Even when allegations were deemed credible, the military often undercounted the toll because victims, unseen by the overhead camera before the strike, remained invisible in the aftermath. Case in point: the 2016 Ramadan bombing near Mosul University that killed five civilians and wounded four. The military reported injuries to two civilians who had been in the pre-strike footage.

Cases Closed

When the military receives an allegation of civilian casualties, it runs through a checklist to determine whether the case merits further inquiry. Most never reach the point of video review. About a quarter of the noncredible cases were summarily closed because they lacked sufficient information or detail, such as a specific location or 48-hour time frame. But more than half were rejected, in some cases erroneously, because the military could find no record of corroborating strikes in the geographic area identified in the allegation — or because there were too many potential matches, and too little detailed information.

That information would be found in official logs maintained by different strike authorities. But The Times found numerous instances in which the logs were incomplete or inaccurate: Often, records show, the coalition knew its logs were flawed.

Frequently, cases were closed because the military said it lacked the information to pinpoint the neighborhood in question. Sometimes that conclusion was rooted in misunderstandings of local custom and culture.

In January 2017, citing insufficient information, an officer quickly closed a case based on social media reports that civilians had been killed in a strike on a funeral in the al Shifaa neighborhood of West Mosul. Fruitlessly, the officer had searched logs for potentially corroborating strikes in the cemetery closest to that neighborhood.

However, as reflected in a graphic video accompanying the initial reports, the strike had not taken place at a cemetery: A thumbnail depicted the entrance to a house. In fact, Muslim funerals are rarely held at cemeteries. What’s more, Muslims bury the dead quickly, and it had been four days since this man, Col. Aziz Ahmed Aziz Sanjari, had died.

The colonel’s death had brought many members of the Sanjari family’s tribe to their home to mourn. It was a sunny afternoon, so more than a dozen people sat outside. They could hear a drone humming above, but were unworried. It was a common occurrence. A few minutes later, the bomb hit. Eleven people were killed, The Times found.

‘Sometimes Bad Things Happen’

Captain Urban acknowledged that, “In some cases our assessment of the numbers of civilian casualties does not always match that of outside groups, and we acknowledge that those numbers may change over time as well.

“We do the best we can, given the circumstances, to understand fully the effects of our operations and the harm done to innocent life. That we sometimes do not always arrive at the same conclusion of outside groups does not diminish the sincerity with which we strive to get it right.”

Several Pentagon studies, rendered in military bureaucratese, have observed some of the failures of accountability. The April 2018 Joint Chiefs of Staff examination of civilian deaths from airstrikes in the Middle East and Africa found that “feedback to subordinate commands on the cause and/or lessons learned from a civilian casualty incident is inconsistent.” The recent Pentagon Inspector General report spoke of “omissions.”

Yet for the most part, these reports do not speak to questions of how airstrikes repeatedly go wrong.

Mr. Lewis, the co-author whose efforts to analyze the assessments in aggregate were excised from the Joint Chiefs’ study, said the report instead relied primarily on interviews with assessment officers. They were able to detect certain patterns — especially casualties from secondary explosions and from people entering the target frame after a weapon’s firing — but few of the systematic reasons behind the bulk of civilian deaths.

The Times asked him why the military would develop such intricate procedures to prevent civilian casualties, and then assess them, but not prioritize documenting or studying causes and lessons learned. Not only does the system provide legitimacy for the military’s actions, he said; it also allows the United States to boast of a process that is a global model of accountability.

The former high-level American official in the campaign against ISIS said the procedures served an additional purpose — to provide a “psychological veneer” for the people involved: “We did the process. We did what we needed to do. Sometimes bad things happen.”

He said that after returning from his post, anguished by what he had seen, he had started therapy. He pointed to Raqqa, rendered a necropolis by American-led airstrikes, and compared it to the ruins of Aleppo, which was bombed by the Russians without the American military’s sophisticated considerations of proportionality — the collateral damage estimates, no-strike lists or rules of engagement.

“Eventually I stopped saying that this was the most precise bombing campaign in the history of warfare,” he said. “So what? It doesn’t matter that this was the most precise bombing campaign and the city looks like this.”

In Afghanistan

All the boys and men of Band-e-Timor knew that when the Toyota Hiluxes came, you should run for your life.

People called them wegos. At the wheel were Afghan paramilitary forces who usually set out on full-moon nights at the fork in the road before Lashkar Gah, charging through the village of Barang straddling the Kandahar-Helmand border and into other parts of Band-e-Timor, “capturing everyone: old men, young men, everyone,” said a resident named Matiullah.

It did not matter if you were not Taliban, people said. If you were male, the Afghan forces would arrest you, simply to collect a bounty for your release. If you were old or feeble, the price was just over $500; a man in his prime would fetch twice that. “You would have to sell your cow or your land to get your relatives released,” said Rahmatullah, a village resident. Often, it was the poorest who would run.

On the night of Jan. 31, 2018, the moon was especially bright. The wegos, as usual, came accompanied by what villagers said were American aircraft. Hidayatullah, a driver by profession, three days from marrying, knew he could not afford the bounty and the wedding, so he drove out into the desert. Then an airstrike found him, said Matiullah, who is his cousin. Dozens of other civilians, mistaken for Taliban as they fled on foot and motorbike across Band-e-Timor, died in the raid as well.

The August drone strike in Kabul that killed an Afghan aid worker and nine of his relatives grabbed the world’s attention. But most American airstrikes in Afghanistan took place far from the cities, in remote areas where cameras were not filming, mobile lines were often cut and the internet was nonexistent.

America’s longest war was, in many ways, its least transparent. For years, these rural battlefields were largely off-limits to American reporters. But after the Taliban returned to power in August, Afghanistan’s hinterlands opened up.

The Times arrived in Barang a little over a month later, visiting 15 households in this hamlet of mud homes and farmland, and also interviewing tribal elders and others across Band-e-Timor. Most said they had never spoken to a journalist before. The accounts they gave — consistently and reliably, in hourslong interviews — help explain how America lost the country, how its war of airstrikes and support of corrupt security forces paved the way for the Taliban’s return.

On average, each household lost five civilian family members. An overwhelming majority of these deaths were caused by airstrikes, most during wego raids. Many people admitted they had relatives who were Taliban fighters, but civilians accounted for most of those lost:

A father killed in an airstrike while running for the forest. A nephew killed as he slept with his flock of sheep. An uncle shot by American soldiers as he went to the bazaar to buy okra for dinner.

At the sound of helicopters, Hajji Muhammad Ismail Agha’s sons had bounded for the desert. The “foreign helicopters” fired on them. One son, Nour Muhammad, was killed; the other, Hajji Muhammad, survived. “How could the planes tell the difference between a civilian and a Taliban?” the father asked. “He was killed just a little far from here. I watched it happen.”

None of these incidents were mentioned in Pentagon press releases. Few were tallied in United Nations counts. So isolated from the Afghan government were residents that when asked for their loved ones’ death certificates, they asked where they might obtain them. Instead, to verify deaths, The Times visited tombstones, in graveyards littered across the desert.

Reporting was contributed by Lila Hassan, Momen Muhanned, Jeff Parrott, Abbie Cheeseman, Hiba Yazbek and Abdul Hadi Patmal. Produced by Michael Beswetherick and Tala Safie.

By Azmat Khan

Photographs by Ivor Prickett

1. “No civilian presence”

For Ali Fathi Zeidan and his extended family, West Mosul was in 2016 still the best of many bad options. Their longtime home in a nearby village, Wana, had been taken by ISIS, then retaken by Kurdish pesh merga forces, and — as if that were not enough — it stood just seven miles below the crumbling Mosul Dam, which engineers had long warned might soon collapse, creating a deluge that would kill everyone in its path. The family had avoided the camps for internally displaced people, where they would have faced a constant risk of separation, and found their way instead to the city, to a grimy industrial neighborhood called Yabisat. They moved into a storage facility, divided it up into separate rooms, brought in a water tank, built a kitchen and a bathroom. Though ISIS had taken Mosul, parts of the city were still relatively safe. Now it was home.

Family was everywhere. Zeidan’s daughter Ghazala was married to a man named Muhammad Ahmed Araj, who grew up in the neighborhood. Araj’s brother, Abdul Aziz Ahmed Araj, lived nearby in a small, crowded apartment. Zeidan’s other daughter moved into an apartment on the other side of Mosul with her husband and their six children, but one of them, 11-year-old Sawsan, preferred to spend her time across town in Yabisat: She was attached to her grandparents and loved playing with her cousins.

Sawsan had been staying with her grandparents for a week when the whole family sat down to dinner on March 5, 2016. All told, there were 21 people around the table. None of them knew that their Iraqi neighborhood was at that moment in the cross hairs of the American military.

Weeks before, Delta Force commandos had captured a high-ranking operative in ISIS’ burgeoning chemical-weapons program, and the information he provided interrogators led military officials to a chemical-weapons production plant in Yabisat; observers had been studying the site for weeks, by way of surveillance flights.

On March 2, military officials presented their findings for validation, as part of the Pentagon’s “deliberate targeting” process, which — as opposed to the rapid process of targeting in the heat of battle — required vetting at multiple levels and stages across the U.S.-led coalition. It had all the makings of a good strike. Unlike with so many other targets, military officials had human intelligence directly from the enemy and video surveillance that showed clear target sites.

They had also concluded that there was no civilian presence within the target compound. Though the surveillance video had captured 10 children playing near the target structure, the military officials who reviewed this footage determined the children would not be harmed by a nighttime strike because they did not live there: They were classified as “transient,” merely passing through during daylight hours.

But as investigators later documented, during the target-validation process one U.S. official disputed this conclusion: A “representative” with the United States Agency for International Development said that the children and their families most likely lived at or around the target compound. In the current environment, she argued, parents would be unlikely to let their children stray far from home. In her view, the determination that there was “no civilian presence” at the target was wrong, and authorizing the strike could lead to the deaths of these children and their parents and families. Military officials dismissed her concerns and authorized the strike.

Three days later, on the evening of March 5, Abdul Aziz heard the explosions, maybe a dozen in all. They came from the direction of his brother’s house. He wanted to see what happened, but because bombings were often accompanied by a second round of missiles, he waited. Later, when he approached the block, he saw the flames and fire consuming what was once his brother’s home. “The place was flattened,” he told me when I first met him, nearly four years later. “It was just rocks and destruction. There was fire everywhere.” They returned at dawn, with blankets to carry the dead. “We searched for our relatives,” he told me, “picking them up piece by piece and wrapping them.”

Across town, Ali Younes Muhammad Sultan, Sawsan’s father, heard the news from his brother. Everyone at the dinner had been killed: Zeidan and his wife, Nofa; Araj, Ghazala and their four children; Zeidan’s adult son Hussein, Hussein’s wife and their six children; Zeidan’s adult son Hassan, Hassan’s wife and their two children; and Sawsan, their own beloved daughter. Sultan and his wife went to the hospital where Sawsan’s remains were taken.

“If it weren’t for her clothes, I wouldn’t have even known it was her,” he later told me. “She was just pieces of meat. I recognized her only because she was wearing the purple dress that I bought for her a few days before. It’s indescribable. I can’t put it into words. My wife — she didn’t even know whether to go to her daughter, or the rest of the family first. It is just too hard to describe. We’re still in denial and disbelief. To this day, we cannot believe what happened. That day changed everything for us.”

2. “Pattern of life”

In the immediate aftermath of the strike, Defense Department officials lauded it as an intelligence coup. But doubts quickly began to surface. A series of ISIS videos taken at the hospital and the strike site was posted online, showing the burned and bloody corpses of children. The coalition opened a civilian casualty review.

The Pentagon’s review process is one of the few, if indeed not the only, means by which the U.S. military holds itself to account with regard to civilian casualties as it executes its air wars. The coalition has conducted at least 2,866 such assessments since the air war against ISIS in Iraq and Syria began in August 2014, but little more than a dozen of the resulting reports have ever been made public until now. Instead, each month, the U.S.-led coalition publishes a summary report, often a series of sentence-long synopses of the findings with little more than the date of the allegation, the general location and what the assessment concluded: that the allegation is “credible” — that is, military investigators deemed it “more likely than not” that an airstrike caused civilian casualties — or that it is “noncredible.”

As I previously reported in The Times, over the past three years, I obtained more than 1,300 of these credibility assessments through the Freedom of Information Act. The reports cover allegations surrounding airstrikes that took place between September 2014 and January 2018. What I saw after studying them was not a series of tragic errors but a pattern of impunity: of a failure to detect civilians, to investigate on the ground, to identify causes and lessons learned, to discipline anyone or find wrongdoing that would prevent these recurring problems from happening again. It was a system that seemed to function almost by design to not only mask the true toll of American airstrikes but also legitimize their expanded use.

Capt. Bill Urban, a spokesman for U.S. Central Command, said the Pentagon worked diligently to prevent the loss of innocent life. “Mistakes do happen,” he said, “whether based upon incomplete information or misinterpretation of the information available. And we try to learn from these mistakes.” But he contested the idea that the Pentagon acted with impunity, noting that “the lawfulness of a military strike is judged upon the information reasonably available to the striking forces at the time of the decision to strike.”

The documents reveal how unreliable that information often was. “White bags” of “ammonium nitrate” at a “homemade explosives factory” were most likely bags of cotton at a gin. A supposed ISIS headquarters was the longtime home of two brothers and their wives and children. An “adult male associated with ISIS”’ was actually an “elderly female.” A man with a weapon “on his left shoulder” actually had no weapon. Males on five motorcycles driving “quickly” and “in formation” — displaying the “signature” of an imminent attack — were just guys on motorbikes. A “heavy object” being dragged into a building was in fact a child.

The documents also offer a window into the process by which strikes are authorized and examined after the fact. The Pentagon’s assessment of what happened at Yabisat, for instance, makes clear that one official who reviewed the intelligence, the U.S.A.I.D. “representative,” warned that there could be civilian casualties. But it nonetheless states that “intelligence associated with the target did not reveal civilian pattern of life” at the target and that video taken before the strike did not reveal “any obvious sign of human activity” in the vicinity. (A spokesman for U.S.A.I.D. declined to comment and referred questions about the case to the Pentagon.) The report also found that the Yabisat strike “fully complied” with the law of war and even “went beyond what is required in terms of harm mitigation” by being conducted at night. Finally, the report recommended that a full investigation be conducted into the “target development and intelligence process” used to determine the “pattern of life” of civilians.

But the records can show us only so much. They tell us what the air war looked like from above, to the officials carrying it out. I knew that to fully understand what was happening, I also needed to see it from the ground. That is the subject of this article. I have spent the past five years traveling throughout the theaters of war in Iraq, Syria and Afghanistan, trying to gain a clear picture of the ground reality created by the air campaign. Starting in 2016, as the U.S. effort against ISIS intensified, I was in cities and towns including Mosul and Hawija, Raqqa and Tokhar. In 2019, as airstrikes occurred at a record pace in Afghanistan, I was meeting families from Helmand, Kandahar and Nangarhar, who gave testimony of night raids and airstrikes that turned even supporters of the embattled Afghan government away.

On the ground, I found a pattern of life that was very different from the one that the military described in its credibility assessments, and documented death rates that vastly exceeded U.S. Central Command’s own numbers. I also came away with a grim understanding of how America’s new high-tech air war looks to civilians who live beneath it — people in Syria, Iraq and Afghanistan trying to raise families, earn a living and stay away from the fighting as best they can. For them, the sight of aerial surveillance drones patrolling the sky overhead is common. It might even provide comfort, suggesting that they were being carefully observed before any action was taken. But they also have come to understand that on occasion, and with no warning, a bomb might pierce the sky, inexplicably targeting their homes, killing their families and neighbors in a terrifying instant.

And they knew that if this were to happen, it was unlikely anyone would ever tell them why.

3. “Extraordinary technology”

In recent decades, the United States has fundamentally transformed its approach to war, replacing American troops on the ground with an arsenal of aircraft directed by controllers sitting at computers, often thousands of miles away. This transformation reached full force in the final years of the Obama administration, amid the deepening unpopularity of the forever wars that had claimed the lives of more than 6,000 American service members. Fewer American troops on the ground meant fewer American deaths, which meant fewer congressional hearings about the progress of the wars, or lack thereof. It also meant fewer reporters paying attention to the impacts of the war effort on the local civilian population. If America could precisely target and kill the right people while taking the greatest possible care not to harm the wrong ones, then those on the home front would have little cause for concern.

From Iraq and Syria to Somalia and Afghanistan, air power allowed coalition forces to take territory from ISIS and the Taliban, and drone strikes provided a means to engage Al Qaeda, Al Shabab and Boko Haram in areas not declared as official battlefields. Military officials touted the precision of these campaigns, based on meticulously gathered intelligence, technological wizardry, carefully designed bureaucratic hurdles and extraordinary restraint. By April 2016, the Pentagon was reporting that American airstrikes in Iraq and Syria had killed 25,000 ISIS fighters, while resulting in the deaths of just 21 civilians. “With our extraordinary technology,” President Barack Obama said that year, “we’re conducting the most precise air campaign in history.”

At the time, I had just finished an investigation into the U.S. government’s claims about the schools it had built in Afghanistan, and I knew that there was often a divergence between what officials say and the reality on the ground. The numbers of civilian casualties given by the coalition seemed hard to believe. So I decided to travel to the sites of some airstrikes and see what I could find out.

In August 2016, coalition forces hit Qaiyara, a suburb about 45 miles south of Mosul, with multiple strikes, freeing it from ISIS control, and in the immediate aftermath, the Pentagon did not acknowledge a single civilian death. I arrived in Qaiyara a little over a month after the strikes had stopped. The air around the town was still thick with black smoke — ISIS fighters had set some oil wells ablaze before retreating north toward Mosul. In the center of Qaiyara, the destruction was absolute. Almost every major building or significant piece of city infrastructure had been hit — the bridges, the water sanitation plant, the railway station, the furniture market, the bazaar. At the remains of Qaiyara’s sloping soccer stadium, I saw children use metal sheets as sleds. The residential area was also devastated: On each block, one or two structures had been reduced to rubble.

I stopped to talk to some local people in front of a destroyed home. They knew the family who used to live there. This was the residence of Ali Khalaf al-Wardi and his family, they told me, as they explained what happened. When the Iraqi Army was advancing toward Qaiyara, fleeing ISIS fighters left explosives caches around the city; Ali, believing that one of those caches was in the house next door, immediately began packing up his family to leave. But they didn’t move quickly enough. A coalition airstrike hit the neighbor’s house, bringing down the Wardi family home. Six civilians were killed, including Ali; his 5-year-old son, Qutada; his 14-year-old daughter, Enaas; and his 18-year-old daughter, Ghofran.

After this, I went to the sites of nine other airstrikes in Qaiyara. All were in residential areas. Locals told me that the airstrikes had rained down daily, particularly in the center of the town. These strikes were so continuous that families frequently slept in shifts in case there was a bombing. At least five of the sites I visited had involved civilian casualties, with at least 29 people killed. In many cases ISIS had already evacuated the homes nearby that were the targets.

It was clear from just one reporting trip that there was something very wrong with the coalition’s air war. I teamed up with Anand Gopal, a journalist with a background in statistical research, and together we mapped out a plan to conduct a systematic ground investigation of airstrikes in Qaiyara. In the coming months, I returned again and again, verifying what I had learned. I broadened my research area to include the town of Shura and the Aden district of East Mosul. I identified impact sites, learned how to distinguish airstrikes from other attacks, interviewed loved ones and survivors, collected names and photographs of the dead, analyzed satellite imagery and scoured social media. Our survey grew to include 103 strike sites, and what we found was sobering: One in five of the bombings resulted in a civilian death, a rate 31 times higher than what the coalition was claiming at the time. What’s more, in about half the strikes that killed civilians, we found no discernible ISIS target nearby. The strikes appeared to have been based on poor or outdated intelligence. It’s true that at that point, we were limited in what we could know about the intended target of a strike. I had military sources, and in some cases I was able to interview local informants on the ground. But my ability to discern pre-strike intelligence was constrained by what these sources would tell me.

Soon, however, I gained deeper insight into the targeting process. On one of my trips, I met an Iraqi man named Basim Razzo, who survived a 2015 strike on his East Mosul home that killed his wife, his daughter, his brother and his nephew. U.S. intelligence had identified the Razzo home as a car-bomb factory. Razzo desperately wanted to know why his family had been targeted so precisely, and to clear his name. After learning about his case, I filed a request under the Freedom of Information Act for the civilian casualty assessment related to this strike. To expedite the process, which can sometimes take years, I argued in my request that there was risk of imminent harm to Razzo, because survivors of U.S. bombings can fall under suspicion of ties to enemy groups. Within months, I had a dozen partly redacted pages.

This was the first report I saw, and it was a revelation to me. My hunch that something had gone very wrong had been correct. The Razzos had been monitored for just 95 minutes over the course of several weeks before the target was authorized, and confirmation bias ran rampant. It didn’t matter that, as the report noted, “no overtly nefarious activity was observed.” Whoever analyzed the surveillance footage interpreted the normal activity of the household through an incriminating lens, noting, for instance, that when Razzo or his brother opened the gate to allow a guest to enter, this was consistent with the tactics, techniques and procedures of an ISIS headquarters; or that the apparent absence of women confirmed that this was an ISIS facility (because Mosul was under ISIS control at the time, the women in the Razzo household rarely went out). There was seemingly nothing the Razzos could have done to persuade the people watching them that they were innocent. In the end, the report acknowledged that perhaps the target may have been confused for a compound next door. (Gopal and I wrote about Razzo’s case in a 2017 article for this magazine called “The Uncounted.”)

Seeing the report for Razzo’s case persuaded me that there was much more to know about the conduct of the air war. If this was the process and intelligence used for a deliberately planned airstrike, vetted at the highest levels, what would the intelligence look like in all the tens of thousands of other strikes, many of them carried out much more quickly than this one? I began filing requests for the thousands of credibility assessments of other strikes in which civilians had reportedly been killed.

4. “We were the sacrifice”

While I waited for those requests to work their way through the system, I returned to Iraq. Razzo and his family had lived on the eastern side of Mosul. In early 2018, I returned to investigate the western side of the city. I wanted to survey the Old City in West Mosul the same way I had the eastern side of the city. My intent was to go methodically from door to door, interviewing locals and documenting each impact area I found. But in the Old City, there were hardly any doors left to knock on. Much of the area had been reduced to rubble. To clear space for vehicles to pass, heavy machinery had been brought in to push the scattered concrete blocks, household debris and even some body parts into little hills of wreckage on either side of the roads.

Near the remains of Al Nuri Grand Mosque, I came upon a makeshift cemetery. While the battle was raging, families quickly buried loved ones here, marking the graves with rocks so they could return and properly lay them to rest when it was safe. Several men had gathered, and I asked them about the costs of liberation from ISIS. One of them, Mudhar Abdul Qadir, stepped forward to share his thoughts. He had lived his whole life in Mosul and was furious over what had happened to his hometown. “We were the sacrifice,” he told me. “We paid the price with our bodies.”

Like Qadir, many of the people I met on this side of the city refused to call it liberation. In their eyes, the government in Baghdad and its American partners made a deliberate choice to punish Mosul and its civilians. As evidence, they pointed to the fact that in every major offensive against ISIS, from Ramadi to Falluja, the coalition had mostly allowed ISIS a longtime convention of war: retreat. This enabled the separation of local civilian populations from combatants. As the coalition closed in on ISIS positions in West Mosul in 2017, everyone assumed this would happen again. But it didn’t. This time, there would be no escape, no path to Syria. This would be the end. Around Mosul, Iraqi and militia forces cut off every exit, trapping civilians with combatants as they made their last stand. Faced with the overwhelming asymmetry of air power and certain defeat, ISIS took swaths of the population hostage. As coalition missiles and bombs rained down, they seemed to kill indiscriminately.

There were widespread reports that coalition strikes supporting the campaign to drive ISIS from the city had killed civilians. Qadir wanted me to see the remains of one of them, which destroyed a home where Tariq Khalil Ibrahim Sanjari and his family were sleeping in April 2017. Although it was only a few meters away from where we stood, it took Qadir and me 30 minutes of climbing over wreckage until we reached the blast area. The debris was pushed up so high around the house, it was like peering down into a basement.

Over the next few months, I was able to conduct multiple interviews with those who lived in the house and the neighborhood, and from these I formed an initial picture of what happened. The Sanjari family had rented this house because their own had been damaged during the war. On the night of the strike, 27 people were asleep in five bedrooms. A little after 12:30 a.m., Sanjari’s son Emad Tariq Khalil Ibrahim woke up struggling to breathe and realized he was partly buried under concrete. After removing the blocks on top of him, he found his wife and two sons. He heard a voice and began searching for other survivors. The lower half of his brother Mahmoud Tariq Khalil was pinned under a block of concrete and steel bars. “I don’t know how to describe the moaning sounds he was making,” Emad told me softly. “I started by hugging my brother, and I kissed his forehead. I told him: ‘Don’t worry, you’re going to be OK. We’re going to save you.’ He didn’t say anything. He just moaned.”

Neighbors who came to help spent more than three hours working to free Mahmoud and the others, but the block was too heavy to lift. Emad could feel Mahmoud’s heartbeat slowing, his body temperature going down. He understood what was happening. Emad kissed his brother, said a prayer and left the room.

Using a drill hammer, a metal-cutting tool and a car jack, the neighbors worked until 1 p.m., rescuing survivors and recovering bodies. Then they took the dead, seven in all, for burial. A year later, when I spoke to him, Emad still could not understand what happened. The family heard planes overhead “24 hours a day,” he told me. What were those planes doing if not providing intelligence that dozens of civilians were in this house? ISIS had previously briefly occupied the house next door to this home, he said, but abandoned it about 20 days to one month before the strike. That home did not appear to be hit.

“What I care about the most, more than anything else, is to help prevent what happened to my family from happening to anyone else,” Emad told me. “Can you uncover the truth about why this house was hit?”

5. “Post-strike analysis”

Uncovering the truth was an almost insurmountable task without the documents. By June 2018, Centcom had denied expedited processing for every single request I submitted. So with lawyers from the Reporters Committee for Freedom of the Press, I filed a lawsuit. By early October, the first batch of casualty reports arrived. Among them were documents from 35 strikes in Iraq in which the Pentagon concluded it was credible that civilians had been killed.

I printed out the more than 300 pages of these reports and began marking them up with questions for military sources. The assessments were littered with acronyms and military slang, made all the more incomprehensible by a steady series of redactions. But just as with the initial document I received regarding the strike on Basim Razzo’s home, these records contained much that was revealing about the nature of America’s air war.

In one especially disquieting assessment, I found a chat log of a conversation among military personnel as they carried out an airstrike in Mosul: They talked about it as if it were a video game, with one saying that the area was “poppin’” with targets, before realizing, just as the chat ended, that they could see children. Another assessment described a strike in which the operators chose to drop a disproportionately large weapon so they could save smaller bombs for later use. The resulting explosion took out two civilian vehicles along with the ISIS vehicle they were targeting.

The documents were especially illuminating when combined with independent ground reporting, something the credibility assessments themselves usually did not contain. None of the investigations, I noted as I turned the pages, included the kind of survivor interviews I had been conducting. The closest thing I found was a description in one document of an interview that Special Operations forces conducted with civilians who had recently fled from an area controlled by ISIS, but it appeared that the intent of the questions was to determine possible ISIS targets to strike, not to glean any information about civilian casualties. Still, given the opportunity, the people in the camp spoke up about the airstrike that killed their neighbors.

Perhaps this disconnect between the documents and the reality on the ground was also the reason the Centcom tally of civilian casualties was consistently lower than what I was finding. An example of this was provided by my visit, in late 2018, to Tokhar, Syria, the site of what was reportedly one of the largest mass-casualty events of the war. The Pentagon claimed that the 2016 strike had killed as many as 24 civilians, but some estimates ran much higher than that, possibly higher than 200. That would make the civilian death toll from the Tokhar strike larger than any other from a coalition airstrike during the war.

It had taken months to persuade the Turkish government to provide me permission to cross the border so I could travel to Tokhar. Now that we had clearance, we left Gaziantep, crossed the border into Syria and drove south. We arrived in Tokhar at noon. Everyone we spoke to recalled the incident. They recounted how, as the fighting between the Syrian Democratic Forces and ISIS grew more and more intense, some 200 villagers from homes near the front line trekked to the outer edge of Tokhar and took shelter in four homes, in a place far from the fighting. They assumed they would be safe there, because ISIS had not been near any of the homes.

But on July 19, coalition forces carried out a series of strikes. For each of the four houses, I wrote up the names of families that perished. The details were consistently corroborated by open-source information, local journalists and others. According to the count I made by speaking to survivors, which I verified over the coming months, at least 120 people died in the Tokhar strike. The interviews with survivors were harrowing. More than a dozen showed me debilitating injuries. Some told me that so many people were killed that there weren’t enough young men left to pull the bodies from the rubble. It took nearly two weeks, and even then, some of the victims were never found.

When I returned to Gaziantep that night and opened my email, I found the credibility assessment from the Tokhar strike waiting for me. Though a full investigation into the incident had been conducted, I received just a single page, a cover sheet of sorts that laid out the basics. A dynamic strike had been called in by a Special Operations force — I later learned from another source that it was Task Force 9 — in northern Syria. Members of Task Force 9, which was supporting the Syrian Democratic Forces, had received reports of ISIS fighters traveling in areas that were “devoid of civilians.” Concluding that the fighters were assembling for a counterattack against the S.D.F., the task force destroyed three “staging sites” and five vehicles. They were confident of having killed 85 ISIS fighters, but the assessment team later concluded that between seven and 24 civilians “may have been intermixed” with ISIS fighters. When I received several more mostly redacted pages from this report two years later, they indicated that the basis for this judgment was “post-strike analysis” and “S.D.F. source reporting.”

The divergence between what I saw in Tokhar and what I read in the Pentagon’s official report made me understand that the document trove I was assembling would need to be approached skeptically, and supplemented with reporting on the ground as often as possible. Between more trips to Iraq, Syria and Afghanistan, I filed for additional assessments, while studying the documents that were coming in steadily every other month or so. Although I could have published some records while waiting for others to be processed, to truly do justice to this, I knew I needed to be able to report out a greater number of them on the ground. Publishing a military document only allows you to see through its eyes in the sky — and from everything I had now learned through my years of reporting on America’s air war, that view alone is usually a dangerous one.

6. “To God we belong”

When the pandemic arrived in the spring of 2020, I had to pause my reporting in Iraq. I spent the time carefully assembling and analyzing the documents I had obtained. I hired two research assistants, Lila Hassan and Jeff Parrott, former students in the conflict-reporting course I teach at Columbia Journalism School, to help me build out the database further. Together we developed a plan for resuming my investigations as soon as travel restrictions lifted. By now, I could better understand the assessments I was reading and had much more material to work with. I had filed more FOIA requests, and they were progressing. Thousands more pages were rolling in, much faster than any one person could handle.

By late 2020, I also had a new type of information, one that could improve my on-the-ground reporting: After years of negotiation, U.S. military officials had finally provided Airwars, a British nonprofit, with military coordinates of the impact sites for all of the credible incidents of civilian casualties it had acknowledged. Until then, it was often difficult to figure out the precise location of a strike listed in the Pentagon’s releases or detailed in the assessments. The releases might say that a particular strike occurred “near Mosul,” but this was practically useless. Even after I started receiving documents, precise location data was almost always redacted, as were most maps or images that would allow me to geolocate them. Now, because of Airwars, I had coordinates that purported to be accurate within 100 meters. I could use this data to go to a site where I knew a strike occurred and start asking questions.

By the time I received my second Covid shot, I had developed a methodology for how to do this. Before visiting a credible site, I would analyze the document to identify central details about the allegation, the intelligence, what the military concluded the target was, how it was authorized to be bombed, what was observed and recorded in footage, chats and mission reports, the casualties assessed and other details. Next, I would research the coordinates the military provided. I would analyze that location in historical satellite imagery, both from before the date of the strike and after it, to identify potential impact areas and examine whether anything matched the target description in the document, or whether it was possible that the coordinates were incorrect.

I decided to start with Mosul. I wanted to sample a large number of sites, and this would take time. Mosul was a place where I had developed the kind of reporting network that would enable me to work safely for several months. To prepare for my visit, I hired two students from the University of Mosul’s department of translation, Momen Muhanned and Zainab Alfakheri, and trained them in some of the basic techniques of investigative journalism.

According to the records, there were 90 credible incidents of civilian casualties in and around Mosul. Some I had previously been to, but many were new to me. We started by examining the area of each strike in local crowdsourced mapping tools to understand a little bit more about the neighborhood and its infrastructure. We also examined open-source material about the incident, such as the sources documented by Airwars, and we conducted our own searches for additional materials, such as ISIS propaganda videos documenting the aftermath of the bombing. (ISIS videos were considered fairly reliable in their accounts of civilian casualties, even by groups that opposed them.) We crosschecked my repository of videos of bombings uploaded by the coalition to see whether any were potential matches. We used the Wayback Machine and other internet-archiving sites to locate materials that may no longer be available elsewhere online. I put all these materials together and imported them into an app I could access on my phone in the field.

In early May, I arrived in Mosul and began visiting strike sites with Muhanned and a local security expert. Over the next two months, I was able to investigate 50 sites there. In each case, we began with the military’s official coordinates of the site — even if content in the document or analysis of imagery suggested it may have been incorrect. In some cases, I was able to conclude the coordinates were inaccurate, but other details in the document or ground reporting led me to the actual site. Of the 50 sites I visited in Mosul, I was able to confirm the details of what happened and locate survivors or eyewitnesses in 27 of them.

At the coordinates, I would try to find the impact site. I would introduce myself to people nearby and ask if they knew what happened there, which often depended on whether they lived in the area during the time period in question. I understood that what people told me could be incorrect, whether because they misremembered or because they were not telling the truth, so I did everything I could to reduce the possibility of misinformation. There were several ways I did this. Though I was now going in with a clear picture of what the military said happened, I always kept this to myself at first and took pains not to ask leading questions. I also always sought out multiple perspectives from eyewitnesses. And I made sure that no one ever had advance notice that I was coming. That way, no one could set up interviews or scope out a place ahead of time. For a given site, there was sometimes extensive information from eyewitnesses in open-source materials that I could read ahead of time, but I was scrupulous about not contacting these individuals over Facebook or Twitter before I arrived, because I knew this could lead to a wider awareness that I was coming and potentially bias the work. Evidence could potentially be doctored; stories could be aligned. Meeting people unplanned at the site would give me the most reliable testimony.

My arrival time depended on the neighborhood. For example, if it was a working-class neighborhood, I could go early in the morning and expect to find people out and about. But if it was a more affluent residential area, streets would be empty, and doorbells would go unanswered. Shop owners and workers could often easily recall basic details, although not usually specific dates. I would frame time around major events — “Eid al-Adha in 2017,” or “two weeks before this neighborhood was liberated,” right after ISIS destroyed the tomb of Nabi Yunus. After getting a sense of what occurred, I would narrow down the possibilities. It was important not to ask questions that were too specific, planting details it would be better to confirm unprompted. For example, instead of asking, “Was a man pushing a cart killed here?” I would inquire about the structure nearby and whether it was ever hit in a bombing.

Sometimes the people I interviewed described exactly what the report said analysts had observed in the footage — details that never appeared on social media. For example, a man in Mosul recalled an extremely specific scene: a missile landing across the street and missing an ISIS member in a wheelchair, followed by a second that hit him directly as he was fleeing, wounding children who had come running out of their homes. I did not ask him about the wheelchair, but his precise account gave me confidence to ask him to sit for an in-depth interview.

Consent involved more than just asking people whether they were willing to be interviewed or quoted. I would explain my objectives and told them specifically where their words, faces or voices might appear. I told them about my goal of making the American public more informed about the consequences of our wars. Many of the subjects were eager to help, immediately inviting me inside their homes, where sometimes interviews could last several hours. I prioritized those who had firsthand accounts to offer: eyewitnesses and family members. Sometimes, though, they did not want to talk about what happened. They would say that it was better to forget, that this was God’s plan. Inna Lillahi wa inna ilayhi raji’un — “To God we belong, and to him we shall return” — was a frequent refrain.

Other survivors had questions about the U.S. condolence-payment scheme and wondered in general whether it would be worth it to speak to me. Before interviews, I was always extremely clear that I was only a journalist, not an aid worker or a representative of an NGO. I explained that I could not be an advocate, but I could share their accounts, and — if they wished — I could include their contact information in my correspondence with the U.S. military. To my knowledge, none of them were ever contacted by a civilian casualty-​assessment officer.

7. “An unknown heavy object”

About two weeks after I arrived in Iraq, using the coordinates as my guide, I pulled up to an intersection in the Zanjili neighborhood of Mosul. According to the documents, this was where, in 2017, military officials observed ISIS fighters launching small surveillance drones from the top of a low-level building. They did not observe civilians or people walking by before firing. But just before the moment of impact, “two transient civilians were observed to walk from an adjacent street into the collateral hazard area,” and “one light truck” was seen driving toward the target.

The review concluded that the two pedestrians were killed, but it didn’t indicate that any other civilians were affected. In Zanjili, I asked some men standing by a wall if they knew of any airstrikes in this area, before the liberation. They pointed to the structure across from us and began to describe how ISIS fighters had kicked out a family and moved into one of the three houses there, which was two stories. Every day for three days, they came here and operated the drones, which looked like homing pigeons. Some of the local children were fascinated by the little devices and would gather to watch them fly around. On the third day, the building was struck.

One of the men, Maher Mahmoud, was hit by the blast. When the bomb struck, he told me, he was walking past the house on his way to see a friend who sold cigarettes in secret. The blast was large. Mahmoud could feel shrapnel in the back of his head, but he knew there would be a second one, so he ran down a nearby road to hide under a van. The second blast was even larger than the first. If someone had managed to survive the first, he thought, there was no way he or she had survived the second.

As I continued to ask around, I found more and more people who wanted to share details about what happened in this strike. They listed names of those killed and injured. Some took their clothes off to show me their injuries. Huddled together over my computer in the guest room in a local home, we watched the video ISIS made of the aftermath of the bombing — a video that had gone totally unmentioned in the U.S. military’s assessment — and asked for help confirming the identity of each person it showed. The scenes were grim and hard to watch. Children in backpacks, unrecognizable bodies, a terrified little girl in the hospital with big brown eyes. Someone told me that the girl was named Aseel. Her father had died, and her leg was injured, but she had survived the attack and was living with her mother across town in the Hermat neighborhood on the city’s outskirts.

When we got to the house, there was a large crowd of children outside. The interior was painted pink. Aseel was several years older now, shy but smiling. Her mother allowed us to photograph the deep scar on her left leg. She told me that after the strike, she was stuck in the living room behind a door that wouldn’t open. When she finally budged it, she could hear her children screaming. “I took my three kids and I ran,” she said. “Two of my children were injured.”

It took several days to document the toll of this strike: 10 deaths and seven injuries. There were even more, people told us, but it would involve traveling to other neighborhoods, tracking down people who had moved away. I decided to stop and move on to the next incident.

Sometimes the documents gave me a specific, haunting detail to go in search of. One concerned a U.S. strike on a “declared hostile force” entering a “defensive fighting position” in Ramadi in November 2015. According to the records, operators had observed “enemy personnel” moving between a tree line and a building. A person was seen “dragging an unknown heavy object” into the building. As aircraft were being called away to other targets, they fired on the building. Upon review of the footage after the strike, an official in the command center reported that the heavy object was actually “a person of smaller stature” accompanying someone who was nearly twice as tall. A review was prompted, which acknowledged that this was “how a child would appear standing next to an adult.” The age and gender of this smaller person could not be determined.

In June, I headed to Ramadi. I wanted to know who this smaller person was and what happened. The military provided coordinates for the strike that were purportedly accurate “within 100 meters,” but they seemed to be at odds with the details in the document, which included a satellite image of the general area. Still, I went to the site, a shady grove, where I asked residents whether there had been any airstrikes on homes in the area. People shook their heads. The nearest anyone could recall was by the waterside. I drove around a 400-meter radius, asking more people. Still nothing.

Next, I tried the area in the imagery, about a kilometer away. It was rural farmland bisected by a road, with houses scattered throughout. I knocked on the door of a rather large-looking home, just to ask permission to park there while I trudged through the farmland. A woman answered. She knew nothing about a nearby airstrike but said I should ask her husband, who wasn’t home. I promised to return later and tried a house across the street. The man who answered said a family’s home had been bombed somewhere in the vicinity, but they were not living in the home at the time.

In the house next door, I finally found a lead — a man who said his home had been hit and rebuilt. He had a daughter, a little girl who sat on my lap as we talked. When we walked out through his back garden to the area where the old house used to be, she held my hand. At the site of the old strike, I took some photos. The man told me he was not in Ramadi at the time, but he called his brother, who was. While we waited for the brother to arrive, I chatted with the man and played with his daughter. When the brother showed up, about half an hour later, I learned that the strike on this property took place on a different date. It was not the strike I was looking for.

In the end, this was one of the 25 out of 60 “credible” sites I visited in Iraq and Syria where I could not verify what happened. In this case, there were three likely scenarios to explain why: The “person of smaller stature” could have been the child of ISIS members, in which case I would be unlikely to ever learn anything about the child; he or she might have been part of a civilian family displaced by violence, now living far away, in a camp somewhere; or the location might be somewhere else altogether.

Another problem stemmed from unreliable record-keeping. In the organizational logs that track all strikes (where the military pulled the coordinates it shared with Airwars), a single “strike” could include more than a dozen engagements lumped together. An official within the air campaign explained to me that this approach to accounting was in part due to the overwhelming number of airstrikes the coalition carried out in Iraq and Syria — it was simply too hard to track them all in the main logs. Therefore, in a case where a single “strike” actually contained multiple engagements, accessing the coordinates for each one required a deeper level of data than what was present in the logs. As a result, fair claims by survivors of strikes have often been rejected on the basis that the military had no record of a strike in that area.

This was probably what happened in the case of the bombing I was trying to investigate in Ramadi. According to the documents, it was the 16th of 17 total “engagements” that had taken place “in and around Ramadi” on Nov. 13, 2015. All of them together were considered one strike. There could be 16 other specific sites to investigate around Ramadi. But where exactly? I had no way of knowing. I had no way of finding the small person who had been mistaken for a heavy object. I had no way of knowing if he or she was alive or dead.

8. “Why did you kill them?”

The rules of war serve many purposes, from shaping concerted action out of the chaos of battle to constraining the technological advances that allow military planners to deliver death with almost boundless ease. They also play a psychological role. As one military official who served at a high level in the air war against ISIS told me, the principles that guide decision-making in war are designed to provide psychological comfort to those who must make the decisions.

That same logic could apply to ordinary Americans as well. Why do people consider the wars in Iraq, Syria and Afghanistan just? How can we know that the next wars will be, too? It is nearly certain that the technologies developed during these past wars will be put to use in the next conflict. Knowing that the American military planners in charge of our new high-tech systems of air war are governed by commitments to specific principles can provide us comfort in the humanity and morality of our government’s actions.

In my quest to understand why American bombs landed where they did, though, I often found myself in the uncomfortable position of having to explain how these principles actually played out in practice.

If there was a single moment in which I most struggled with this, it was during an interview in June in Al Tanak, a neighborhood on the outskirts of West Mosul. In 2017, military planners identified a home where ISIS fighters slept — a “bed-down location” — based on reports from five sources. The structure was a residential home and thus had been on a restricted targeting list, until the military concluded it was exclusively used by ISIS. Its protected status was removed, and the target was approved.

But on the day of the planned strike, observers saw something they did not expect: three children on the roof. Their presence raised questions about whether the house was being exclusively used by ISIS, and the strike package was returned to the targeting team for further evaluation. The next day, the target’s “casualty estimation worksheet” was updated: Three children, who probably lived there, were also included.

This was not an error. According to U.S. rules of engagement, military planners can knowingly kill civilians, including children, if the anticipated casualty rate is not excessive in relation to the anticipated military advantage of destroying the strike target.

Observers continued to watch the house, and now they thought they saw something even more dangerous: a “[redaction] coming from the target” led them to believe ISIS was manufacturing weapons there. Not long after weapons were fired, children were seen south of the compound, and a screener watching the video observed that “one possible child was carried out of the strike location,” loaded into a vehicle and driven to a medical facility. They concluded that three ISIS members were killed, two were wounded and one child was injured.

Not long afterward, an ISIS-linked media agency tweeted that 11 civilians were killed in Al Tanak. The Pentagon investigators acknowledged this but also noted that the “degree may be significantly exaggerated given the source.”

How many died? Last June, with the document loaded on my phone and curious children swarming us, Momen and I knocked on doors on a residential block in Al Tanak. I asked a woman if there had been any bombings that killed civilians in the area. She pointed down the street and told me that 11 members of a family had been killed by an explosion, and only one member had survived, a little girl. But the house next door was also hit, the woman said, and another girl was injured by the bombing. I could talk to them and find out more.

I thanked the woman and went to the green-​gated house a little farther down the road. There, Obaid Abdullah Younes and his wife, Nisreen Abdullah Younes, invited me inside to talk. Nisreen said Fatima, her daughter, was 3 months old when the bomb hit. “Something fell on her head and made her like this,” her mother told me, gesturing toward a 4-year-old girl who was writhing on the floor, her mouth open. As we talked, curious neighbors wandered in without knocking. They all had different theories about why the house next door was hit. It made no sense. The family had lived in the neighborhood for 40 years. Everyone knew them, they said, and they had nothing to do with ISIS. Maybe it was the ISIS bunk house across the street, even though it had been vacated before the strike and untouched. Or maybe it was the motorcycle that was parked outside, or the engine in front of the house stripped for spare parts. Whatever the theory, no one would say a bad word about the family of Muhammad Ahmed Muhammad Muhammad Mousa.

I asked if they knew where the lone survivor, whose name was Rahaf, had gone. Nisreen said yes: She had gone to live with her grandparents, whose house was only a short drive away. One of their children rode in the car with Momen and me to show us where.

Mousa’s sister Katbeeah Ahmed had an immediate warmth about her. She smiled, invited us into her guest room and gave us ice water in a delicate bowl. I began my interview in my normal manner, telling her that I was a journalist trying to understand what happened in incidents that harmed civilians. I said I wanted to ask her questions about difficult things, but that if I asked anything that was hurtful, she could say no at any time. She said to ask whatever I wanted.

Her daughter Esraa was married to her brother’s son, and so Katbeeah had lost more than just her brother, she said. She lost her daughter, who was her best friend, and grandchildren, and nephews and nieces. After the bombing, neighbors heard the sound of what they thought was a cat screaming from underneath the rubble. They lifted the blocks and found Rahaf.

That night, Katbeeah told me, Mousa and the family had planned to come to her house for a meal. They were poor and often could barely afford to eat. They died without eating dinner, she said. Her brother formerly worked as a guard in Badush prison but left the job after ISIS took over. Maa’n, her daughter’s husband, was a nursing student but had to leave school as well. Mousa was excited for liberation. He kept a TV hidden in his house, against ISIS strictures, and watched the Iraqi news eagerly for updates on the advancing Iraqi Army.

I asked her why she thought the bomb hit her brother’s home.

Katbeeah was certain it was some kind of mistake. A random strike gone wrong. Still, she and the other family members who came to listen to the interview shared the different theories they and neighbors had speculated on. An ISIS truck was parked underneath a tree or had been driving by. Maybe they meant to target that. On their roof was a tanoor, a mud oven to make bread. It used a cheaper oil that burned deeply. Maybe they saw its heat from the sky.

I wondered if the redacted item that was “coming from the target” was smoke or a heat signature from the oven.

Nearly an hour and a half into our conversation, I told her about the document. I summed up the initial description: that they believed the house was being used by ISIS for military purposes. Then I told her that, before striking, they saw three children on the roof.

Katbeeah’s face changed. The children would go up on the roof when they got cold, she said quietly. It was January. The house did not have gas. On the roof they could warm up under the sun.

I described how after seeing the children, the target was re-evaluated, and they saw something coming from the house that made them believe it was a weapons manufacturing facility.

She encouraged me to go out into the neighborhood and ask about her family. “Everyone will tell you the exact same thing,” she said. “It’s impossible.”

I asked Katbeeah what she would want to tell the people who wrote the document and who did this bombing.

“I am on fire now,” she said, her voice robbed of all its signature warmth. “Why did you kill them? They were innocent. They didn’t do anything.” Now she was weeping. “They were turned into just flesh. Their house wasn’t suspicious at all. I ask now, I want to know the reason. There wasn’t any manufacturing facility.”

Katbeeah was sobbing. I apologized for asking. My own voice was cracking now. She told me that she was grateful, that they wanted to know this, that she was happy I was investigating their deaths, that she has never forgotten them.

“I can still see their shadows in front of me,” she said.

I told Katbeeah I wanted to ask her one last thing. I described how military observers believed that the strike was acceptable because the military advantage gained by eliminating an ISIS weapons manufacturing facility would be worth killing the children. What did she think of the decision?

“But they didn’t gain any advantage,” she said. “The only thing they did is they killed the children.”

Lila Hassan, Momen Muhanned, Jeff Parrott, Ali Uthman, Abdullah Abdelqader Ali, Mahmood Zaki, Abbie Cheeseman, Hiba Yazbek and Zainab Alfakheri contributed reporting. Additional design and development by Jacky Myint.

By Azmat Khan, Lila Hassan, Sarah Almukhtar and Rachel Shorey

The New York Times is making public hundreds of the Pentagon’s confidential assessments of reports of civilian casualties resulting from U.S.-led airstrikes in Iraq and Syria. The documents lay bare how the air war has been marked by deeply flawed intelligence, rushed and often imprecise targeting, and the deaths of thousands of innocent civilians, many of them children.

The documents were obtained through Freedom of Information requests beginning in March 2017 and subsequent lawsuits filed against the Defense Department and the U. S. Central Command. To date, The Times has received more than 1,300 reports examining airstrikes in Iraq and Syria from September 2014 to January 2018, more than 5,400 pages in all.

Independent reporting by The Times has closely matched much of the basic information from the documents, but it also found important discrepancies and oversights in some instances, including the location of a strike or the number of people killed or injured. Despite the inaccuracies, the documents serve as an important record for researchers seeking to understand the Pentagon’s internal processes.

By Christoph Koettl, Evan Hill, Matthieu Aikins, Eric Schmitt, Ainara Tiefenthäler and Drew Jordan

A week after a New York Times visual investigation, the U.S. military admitted to a tragic mistake in an Aug. 29 drone strike in Kabul that killed 10 civilians, including an aid worker and seven children.

By Christoph Koettl, Evan Hill, Matthieu Aikins, Jim Huylebroek, Ainara Tiefenthäler, Dmitriy Khavin and Eric Schmitt

Even as the military apologized for killing 10 civilians by mistake in a drone strike in Kabul, it insisted its target had stopped by an ISIS “safe house.” The New York Times found that the building was actually home to an NGO worker and his family.

The military never conducted an independent investigation into a 2019 bombing on the last bastion of the Islamic State, despite concerns about a secretive commando force.

By Dave Philipps and Eric Schmitt

In the last days of the battle against the Islamic State in Syria, when members of the once-fierce caliphate were cornered in a dirt field next to a town called Baghuz, a U.S. military drone circled high overhead, hunting for military targets. But it saw only a large crowd of women and children huddled against a river bank.

Without warning, an American F-15E attack jet streaked across the drone’s high-definition field of vision and dropped a 500-pound bomb on the crowd, swallowing it in a shuddering blast. As the smoke cleared, a few people stumbled away in search of cover. Then a jet tracking them dropped one 2,000-pound bomb, then another, killing most of the survivors.

It was March 18, 2019. At the U.S. military’s busy Combined Air Operations Center at Al Udeid Air Base in Qatar, uniformed personnel watching the live drone footage looked on in stunned disbelief, according to one officer who was there.

“Who dropped that?” a confused analyst typed on a secure chat system being used by those monitoring the drone, two people who reviewed the chat log recalled. Another responded, “We just dropped on 50 women and children.”

An initial battle damage assessment quickly found that the number of dead was actually about 70.

The Baghuz strike was one of the largest civilian casualty incidents of the war against the Islamic State, but it has never been publicly acknowledged by the U.S. military. The details, reported here for the first time, show that the death toll was almost immediately apparent to military officials. A legal officer flagged the strike as a possible war crime that required an investigation. But at nearly every step, the military made moves that concealed the catastrophic strike. The death toll was downplayed. Reports were delayed, sanitized and classified. United States-led coalition forces bulldozed the blast site. And top leaders were not notified.

The Defense Department’s independent inspector general began an inquiry, but the report containing its findings was stalled and stripped of any mention of the strike.

“Leadership just seemed so set on burying this. No one wanted anything to do with it,” said Gene Tate, an evaluator who worked on the case for the inspector general’s office and agreed to discuss the aspects that were not classified. “It makes you lose faith in the system when people are trying to do what’s right but no one in positions of leadership wants to hear it.”

Mr. Tate, a former Navy officer who had worked for years as a civilian analyst with the Defense Intelligence Agency and the National Counterterrorism Center before moving to the inspector general’s office, said he criticized the lack of action and was eventually forced out of his job.

The details of the strikes were pieced together by The New York Times over months from confidential documents and descriptions of classified reports, as well as interviews with personnel directly involved, and officials with top secret security clearances who discussed the incident on the condition that they not be named.

The Times investigation found that the bombing had been called in by a classified American special operations unit, Task Force 9, which was in charge of ground operations in Syria. The task force operated in such secrecy that at times it did not inform even its own military partners of its actions. In the case of the Baghuz bombing, the American Air Force command in Qatar had no idea the strike was coming, an officer who served at the command center said.

In the minutes after the strike, an alarmed Air Force intelligence officer in the operations center called over an Air Force lawyer in charge of determining the legality of strikes. The lawyer ordered the F-15E squadron and the drone crew to preserve all video and other evidence, according to documents obtained by The Times. He went upstairs and reported the strike to his chain of command, saying it was a possible violation of the law of armed conflict — a war crime — and regulations required a thorough, independent investigation.

But a thorough, independent investigation never happened.

This week, after The New York Times sent its findings to U.S. Central Command, which oversaw the air war in Syria, the command acknowledged the strikes for the first time, saying 80 people were killed but the airstrikes were justified. It said the bombs killed 16 fighters and four civilians. As for the other 60 people killed, the statement said it was not clear that they were civilians, in part because women and children in the Islamic State sometimes took up arms.

“We abhor the loss of innocent life and take all possible measures to prevent them,” Capt. Bill Urban, the chief spokesman for the command, said in the statement. “In this case, we self-reported and investigated the strike according to our own evidence and take full responsibility for the unintended loss of life.”

The only assessment done immediately after the strike was performed by the same ground unit that ordered the strike. It determined that the bombing was lawful because it killed only a small number of civilians while targeting Islamic State fighters in an attempt to protect coalition forces, the command said. Therefore no formal war crime notification, criminal investigation or disciplinary action was warranted, it said, adding that the other deaths were accidental.

But the Air Force lawyer, Lt. Col. Dean W. Korsak, believed he had witnessed possible war crimes and repeatedly pressed his leadership and Air Force criminal investigators to act. When they did not, he alerted the Defense Department’s independent inspector general. Two years after the strike, seeing no evidence that the watchdog agency was taking action, Colonel Korsak emailed the Senate Armed Services Committee, telling its staff that he had top secret material to discuss and adding, “I’m putting myself at great risk of military retaliation for sending this.”

“Senior ranking U.S. military officials intentionally and systematically circumvented the deliberate strike process,” he wrote in the email, which was obtained by The Times. Much of the material was classified and would need to be discussed through secure communications, he said. He wrote that a unit had intentionally entered false strike log entries, “clearly seeking to cover up the incidents.” Calling the classified death toll “shockingly high,” he said the military did not follow its own requirements to report and investigate the strike.

There was a good chance, he wrote, that “the highest levels of government remained unaware of what was happening on the ground.”

Colonel Korsak did not respond to requests for comment.

Undercounted Tolls

The United States portrayed the air war against the Islamic State as the most precise and humane bombing campaign in its history. The military said every report of civilian casualties was investigated and the findings reported publicly, creating what the military called a model of accountability.

But the strikes on Baghuz tell a different story.

The details suggest that while the military put strict rules in place to protect civilians, the Special Operations task force repeatedly used other rules to skirt them. The military teams counting casualties rarely had the time, resources or incentive to do accurate work. And troops rarely faced repercussions when they caused civilian deaths.

Even in the extraordinary case of Baghuz — which would rank third on the military’s worst civilian casualty events in Syria if 64 civilian deaths were acknowledged — regulations for reporting and investigating the potential crime were not followed, and no one was held accountable.

The military recently admitted that a botched strike in Kabul, Afghanistan, in August killed 10 civilians, including seven children. But that kind of public reckoning is unusual, observers say. More often, civilian deaths are undercounted even in classified reports. Nearly 1,000 strikes hit targets in Syria and Iraq in 2019, using 4,729 bombs and missiles. The official military tally of civilian dead for that entire year is only 22, and the strikes from March 18 are nowhere on the list.

A Secret Task Force

The battle at Baghuz represented the end of a nearly five-year United States-led campaign to defeat the Islamic State in Syria and was a foreign policy triumph for President Donald J. Trump.

At the height of its rule in 2014, the Islamic State controlled an area of Syria and Iraq about the size of Tennessee. A fleet of coalition drones, jets, attack helicopters and heavy bombers hit enemy positions with about 35,000 strikes over the next five years, plowing a path for local Kurdish and Arab militias to reclaim ground.

At the end of the grinding fight, airstrikes corralled the last Islamic State fighters in a scrap of farmland against the Euphrates River near Baghuz. Coalition air power forced thousands to surrender, sparing the lives of untold numbers of Kurdish and Arab allies.

On the ground, Task Force 9 coordinated offensives and airstrikes. The unit included soldiers from the 5th Special Forces Group and the Army’s elite commando team Delta Force, several officials said.

Over time, some officials overseeing the air campaign began to believe that the task force was systematically circumventing the safeguards created to limit civilian deaths.

The process was supposed to run through several checks and balances. Drones with high-definition cameras studied potential targets, sometimes for days or weeks. Analysts pored over intelligence data to differentiate combatants from civilians. And military lawyers were embedded with strike teams to ensure that targeting complied with the law of armed conflict. In combat situations, the process might take only minutes, but even then the rules required teams to identify military targets and minimize civilian harm. At times, when the task force failed to meet those requirements, commanders in Qatar and elsewhere denied permission to strike.

But there was a quick and easy way to skip much of that oversight: claiming imminent danger.

The law of armed conflict — the rule book that lays out the military’s legal conduct in war — allows troops in life-threatening situations to sidestep the strike team lawyers, analysts and other bureaucracy and call in strikes directly from aircraft under what military regulations call an “inherent right of self-defense.”

Task Force 9 typically played only an advisory role in Syria, and its soldiers were usually well behind the front lines. Even so, by late 2018, about 80 percent of all airstrikes it was calling in claimed self-defense, according to an Air Force officer who reviewed the strikes.

The rules allowed U.S. troops and local allies to invoke it when facing not just direct enemy fire, but anyone displaying “hostile intent,” according to a former officer who deployed with the unit numerous times. Under that definition, something as mundane as a car driving miles from friendly forces could in some cases be targeted. The task force interpreted the rules broadly, the former officer said.

The aftermath of that approach was plain to see. A number of Syrian towns, including the regional capital, Raqqa, were reduced to little more than rubble. Human rights organizations reported that the coalition caused thousands of civilian deaths during the war. Hundreds of military assessment reports examined by The Times show the task force was implicated in nearly one in five coalition civilian casualty incidents in the region.

Publicly, the coalition insisted the numbers were much lower. Privately, it became overwhelmed by the volume of civilian casualty claims reported by locals, humanitarian groups and the news media, and a backlog of civilian casualty assessment reports sat unexamined for months, two people who compiled the reports said.

But even when completed, the military teams making those assessments were not equipped to make an accurate count, the former task force officer said, because the personnel doing the counting did not investigate on the ground and often based their findings on how many dead civilians they could definitively identify from aerial footage of the rubble.

Mr. Tate, who wrote a classified report on the shortcomings of the process, said the assessment teams at times lacked training and some did not have security clearances to even view the evidence.

The assessments of the strike process were also flawed, three officials said, because they were done by the units that called in the strikes, meaning the task force was grading its own performance. Rarely did it find problems.

Alarm at the C.I.A.

Human rights groups were not the only ones sounding the alarm. C.I.A. officers working in Syria grew so alarmed over the task force’s strikes that agents reported their concern to the Department of Defense inspector general, which investigated the claims and produced a report. The results of that report are top secret, but the former task force officer, who reviewed the report, said the C.I.A. officers alleged that in about 10 incidents, the secretive task force hit targets knowing civilians would be killed.

The former officer said the report determined that all the strikes were legal.

The inspector general declined to release the report or discuss its findings.

Staff in the operations center in Qatar, who oversaw the air war, also became concerned with task force strikes. Air Force lawyers started keeping a spreadsheet, recording the self-defense justifications the task force used to call strikes, then comparing them with drone footage and other evidence, according to one officer who viewed the data. The evidence appeared to show that the task force was adding details that would legally justify a strike, such as seeing a man with a gun, even when those details were not visible in the footage.

Though a number of officers in the operations center suspected that the task force was including misleading information in the logs to justify strikes, they did not feel they had enough evidence to press the issue, the officer said. That changed on March 18, 2019.

A Fatal Strike

The camp at Baghuz was effectively the Islamic State’s Alamo — a last stand where hard-core militants vowed to fight to the death. For more than a month, they had been trapped in one square mile of burned-over farm fields. Among the makeshift tents, bullet-pocked vehicles and hand-dug bunkers were tens of thousands of women and children. Some were there willingly; some were not.

The coalition had laid siege, hoping to starve the fighters out. In six weeks, 29,000 people, most of them women and children, surrendered. On March 18, drone footage showed the camp still harbored large numbers of people suspected of being fighters and their families.

Coalition drones had scoured the camp 24 hours a day for weeks and knew nearly every inch, officers said, including the daily movements of groups of women and children who gathered to eat, pray and sleep near a steep river bank that provided cover.

What happened on the morning of March 18 is in dispute.

That day Islamic State fighters trapped in the camp launched a predawn counteroffensive, according to Central Command, which oversaw Task Force 9. It said hundreds of Islamic State fighters started firing rifles and grenade launchers and sending forward fighters with suicide vests. The coalition pummeled the fighters with airstrikes — so many that by midmorning the coalition had used all the missiles on its drones. Only one American drone, controlled by the task force, was left in the area, and it was unarmed.

At about 10 a.m., local Syrian forces reported they were under fire and in danger of being overrun, and called for an airstrike, Central Command said. The task force drone tracked a group of fighters as they made their way through the camp to the area where the women and children sheltered.

A 5th Special Forces Group officer in the task force looked at the drone footage and didn’t see any civilians, a task force officer said. But the drone he relied on had only a standard-definition camera. Central Command said there were no high-definition drones in the area that could get a better view of the target.

The Special Forces officer gave the order to fire. With no precision missiles left, the command said, the ground commander called in 500- and 2,000-pound bombs. The strike log classified the strike as self-defense.

In fact, a high-definition drone was available. The task force did not use it. Circling above, it was streaming footage of the same patch of ground to the operations center in Qatar. Because the task force operated at a high level of secrecy, two officers said, the people in Qatar watching the high-definition drone were not aware the task force was about to call in a strike.

Central Command said the task force did not know that the better drone was overhead.

The high-definition drone recorded a very different scene from what was described by Central Command this past week, three people who viewed the footage said. In it, two or three men — not 16 — wander through the frame near the crowd. They have rifles but do not appear to be maneuvering, engaging coalition forces or acting in a way that would seem to justify a self-defense strike with 2,000-pound bombs. A chat log used by analysts who were watching the footage noted the presence of women, children and a man with a gun, but did not mention any active combat, two people who viewed the log said.

The Visual Investigations team at The Times reviewed hundreds of photos, videos and satellite images of the Islamic State camp in Baghuz. The reported strike point lies between two aqueducts, which the team used as reference features to pinpoint the location.

A photograph taken the previous day shows several makeshift tents in the area.

What is not in dispute is that moments after the task force called in the strike, an F-15E attack plane hit the spot with a 500-pound bomb. Five minutes later, when ground forces saw people fleeing the blast site, the F-15E dropped two 2,000-pound bombs on the survivors. The entire attack took 12 minutes.

A Syrian videographer, Gihad Darwish, captured airstrikes in the area matching that description as he filmed from a rocky bluff above the camp. The footage shows that ground troops may not have been able to see the group of civilians.

A Failed Investigation

Defense Department regulations require any “possible, suspected or alleged” violation of the law of armed conflict to be reported immediately to the combatant commander in charge, as well as criminal investigators, the Joint Chiefs of Staff, the secretary of defense and the secretary of the Army.

After viewing the footage, the Air Force lawyer, Colonel Korsak, ordered the units involved to preserve nine pieces of evidence, including video, and reported the strike to his chain of command, according to the email he later sent to the Senate Armed Services Committee staff. He also notified the command of concerns that the unit appeared to be covering up the alleged war crimes violations by adding details to the strike log that would justify a self-defense strike.

He told the committee staff that commanders did not take action.

Coalition forces overran the camp that day and defeated the Islamic State a few days later. The yearslong air war was hailed as a triumph. The commander of the operations center in Qatar authorized all personnel to have four drinks at the base bar, lifting the normal three-drink limit.

Civilian observers who came to the area of the strike the next day found piles of dead women and children. The human rights organization Raqqa Is Being Slaughtered Silently posted photos of the bodies, calling it a “terrible massacre.”

Satellite images from four days later show the sheltered bank and area around it, which were in the control of the coalition, appeared to have been bulldozed.

David Eubank, a former U.S. Army Special Forces soldier who now runs the humanitarian organization Free Burma Rangers, walked through the area about a week later. “The place had been pulverized by airstrikes,” he said in an interview. “There was a lot of freshly bulldozed earth and the stink of bodies underneath, a lot of bodies.”

Concerned that details of the airstrike would be buried as well, Colonel Korsak alerted the Air Force’s version of the F.B.I., the Office of Special Investigations. In an email Colonel Korsak shared with the Senate Armed Services Committee, a major responded that agents probably would not look into it, saying the office typically investigated civilian casualty reports only when there was “potential for high media attention, concern with outcry from local community/government, concern sensitive images may get out.”

The Air Force Office of Special Investigations declined to comment.

Colonel Korsak again pressed his chain of command to act, informing his command’s chief legal officer in a memo in May 2019 that regulations required an investigation. He later told the Senate committee’s staff that his superiors did not open an investigation.

“The topic and incidents were dead on arrival,” he wrote. “My supervisor refused to discuss the matter with me.”

The chief legal officer, Colonel Matthew P. Stoffel, did not respond to requests for comment.

The task force finished up a civilian casualty report on the strike that month and determined that four civilians were killed. But two and a half years later, on the military’s website for its campaign against the Islamic State, known as Operation Inherent Resolve, the military still publicly lists the case as “open.”

A Report Buried

Unwilling to let the issue drop, Colonel Korsak filed a hotline complaint with the inspector general’s office in August 2019.

A four-person team in the office was already looking into shortcomings in the civilian casualty reporting processes in Syria and quickly set up an interview in a secure setting. After reviewing the high-definition footage and interviewing Colonel Korsak, the team, which included Mr. Tate, told superiors in the inspector general’s office that the allegation of a war crime was “extremely credible.”

“When he came to us, he wanted to make it very clear he had tried everything else first,” Mr. Tate said. “He felt that the I.G. hotline was the only option remaining.”

But like the Air Force lawyer’s earlier effort, Mr. Tate’s team soon hit roadblocks. Central Command was slow to turn over evidence, he said. Mr. Tate obtained video from several drones flying over Baghuz that day, but could not locate the footage from the task force drone that called in the strike.

The inspector general’s office received a second complaint on the hotline about the strike, a spokeswoman said, but Mr. Tate said his team was never told.

Mr. Tate studied the task force’s casualty report, but it didn’t match what he saw on video. The civilian deaths stated in the report were “an impossibly small number,” he said.

The final section of the casualty report was reserved for the legal opinion. In one version of the report that Mr. Tate was sent by the staff at Operation Inherent Resolve, the Baghdad-based military command overseeing operations in Iraq and Syria, a task force lawyer and an operations officer wrote that a violation of the law of armed conflict may have taken place. In another copy that came from Central Command, he said, that opinion had been removed.

Mr. Tate could find no evidence that the Joint Chiefs, the defense secretary or criminal investigators had been alerted, as required.

Within days of interviewing Colonel Korsak, Mr. Tate’s team took their findings to supervisors and told them the office was required to alert those officials and criminal investigation agencies. Mr. Tate said his supervisors took no action. The team pressed leaders numerous times over the next several months, and in January 2020, Mr. Tate’s team leader drafted a memo that would formally alert authorities. It only needed to be signed by the deputy inspector general overseeing the team. Mr. Tate said the supervisor did not sign it.

In the months that followed in 2020, the team finished its report on broader issues in the civilian casualty reporting process, but as it went through the editing and approval process, which included comments from Central Command, all mentions of the Baghuz strike were cut.

Mr. Tate became increasingly pointed in criticizing the leadership of the inspector general’s office. In October 2020, he said he was forced out of his position and escorted from the building by security.

The inspector general report on civilian casualties was formally released this spring to select members of Congress and the military with the proper security clearances. The office refused to release a public copy or discuss the classified findings, but acknowledged it did not mention Baghuz.

A spokeswoman for the inspector general’s office disputed Mr. Tate’s account. She said that it alerted the proper authorities at Central Command shortly after receiving the first hotline complaint in 2019. The spokeswoman said the office also notified criminal investigators about the strike in October 2020, 14 months after receiving the hotline call — around the time that Mr. Tate was terminated.

A spokeswoman for the office said a new evaluation of Special Operations Command’s adherence to the law of war was expected be completed this month, and that it would include the Baghuz strike. That report will also be classified.

After leaving the office, Mr. Tate refused to give up. He contacted the Senate Armed Services Committee in May and sent a 10-page letter describing the strike and what he viewed as a “systematic failure” on civilian casualty reporting. The committee then contacted Colonel Korsak, who replied with a detailed email.

When asked by The Times about the March 2019 strike, Chip Unruh, a spokesman for Senator Jack Reed, Democrat of Rhode Island and the chairman of the Senate Armed Services Committee, declined to comment on details of the incident, about which the Central Command has briefed the committee.

He did, however, provide a broader assessment: “When tragic errors occur on the battlefield, the United States, as the leader of the free world, has an obligation to be transparent, take responsibility, and do everything we can to learn from and prevent future mistakes.”

Mr. Tate waited for months for the committee to call back and give him an indication that it was actively looking at the case. This week, he said with a sigh that he was still waiting.

Azmat Khan, Christoph Koettl and Haley Willis contributed reporting. Drew Jordan contributed production.

An American strike cell alarmed its partners as it raced to defeat the enemy.

By Dave Philipps, Eric Schmitt and Mark Mazzetti

A single top secret American strike cell launched tens of thousands of bombs and missiles against the Islamic State in Syria, but in the process of hammering a vicious enemy, the shadowy force sidestepped safeguards and repeatedly killed civilians, according to multiple current and former military and intelligence officials.

The unit was called Talon Anvil, and it worked in three shifts around the clock between 2014 and 2019, pinpointing targets for the United States’ formidable air power to hit: convoys, car bombs, command centers and squads of enemy fighters.

But people who worked with the strike cell say in the rush to destroy enemies, it circumvented rules imposed to protect noncombatants, and alarmed its partners in the military and the C.I.A. by killing people who had no role in the conflict: farmers trying to harvest, children in the street, families fleeing fighting, and villagers sheltering in buildings.

Talon Anvil was small — at times fewer than 20 people operating from anonymous rooms cluttered with flat screens — but it played an outsize role in the 112,000 bombs and missiles launched against the Islamic State, in part because it embraced a loose interpretation of the military’s rules of engagement.

“They were ruthlessly efficient and good at their jobs,” said one former Air Force intelligence officer who worked on hundreds of classified Talon Anvil missions from 2016 to 2018. “But they also made a lot of bad strikes.”

The military billed the air war against the Islamic State as the most precise and humane in military history, and said strict rules and oversight by top leaders kept civilian deaths to a minimum despite a ferocious pace of bombing. In reality, four current and former military officials say, the majority of strikes were ordered not by top leaders but by relatively low-ranking U.S. Army Delta Force commandos in Talon Anvil.

The New York Times reported last month that a Special Operations bombing run in 2019 killed dozens of women and children, and that the aftermath was concealed from the public and top military leaders. In November, Defense Secretary Lloyd J. Austin III ordered a high-level investigation into the strike, which was carried out by Talon Anvil.

But people who saw the task force operate firsthand say the 2019 strike was part of a pattern of reckless strikes that started years earlier.

When presented with The Times’s findings, several current and former senior Special Operations officers denied any widespread pattern of reckless airstrikes by the strike cell and disregard for limiting civilian casualties. Capt. Bill Urban, a spokesman for the military’s Central Command, which oversees operations in Syria, declined to comment.

As bad strikes mounted, the four military officials said, Talon Anvil’s partners sounded the alarm. Pilots over Syria at times refused to drop bombs because Talon Anvil wanted to hit questionable targets in densely populated areas. Senior C.I.A. officers complained to Special Operations leaders about the disturbing pattern of strikes. Air Force teams doing intelligence work argued with Talon Anvil over a secure phone known as the red line. And even within Talon Anvil, some members at times refused to participate in strikes targeting people who did not seem to be in the fight.

The four officials worked in different parts of the war effort, but all interacted directly with Talon Anvil on hundreds of strikes and soon grew concerned with its way of operating. They reported what they were seeing to immediate superiors and the command overseeing the air war, but say they were ignored.

The former Air Force intelligence officer, who worked almost daily on missions from 2016 to 2018, said he notified the main Air Force operations center in the region about civilian casualties several times, including after a March 2017 strike when Talon Anvil dropped a 500-pound bomb on a building where about 50 people were sheltering. But he said leaders seemed reluctant to scrutinize a strike cell that was driving the offensive on the battlefield.

Every year that the strike cell operated, the civilian casualty rate in Syria increased significantly, according to Larry Lewis, a former Pentagon and State Department adviser who was one of the authors of a 2018 Defense Department report on civilian harm. Mr. Lewis, who has viewed the Pentagon’s classified civilian casualty data for Syria, said the rate was 10 times that of similar operations he tracked in Afghanistan.

“It was much higher than I would have expected from a U.S. unit,” Mr. Lewis said. “The fact that it increased dramatically and steadily over a period of years shocked me.”

Mr. Lewis said commanders enabled the tactics by failing to emphasize the importance of reducing civilian casualties, and that Gen. Stephen J. Townsend, who commanded the offensive against the Islamic State in 2016 and 2017, was dismissive of widespread reports from news media and human rights organizations describing the mounting toll.

In a telephone interview, General Townsend, who now heads the military’s Africa Command, said outside organizations that tracked civilian harm claims often did not vet allegations rigorously enough. But he strongly denied that he didn’t take civilian casualties seriously. “There’s nothing further from the truth,” said General Townsend, who added that as commander he ordered monthly civilian casualty reports in Iraq and Syria be made public. He blamed any civilian casualties on “the misfortunes of war” and not because “we didn’t care.”

With few Americans on the ground, it was difficult to get reliable counts of civilian deaths, according to Gen. Joseph L. Votel, the head of the military’s Central Command at the time, and General Townsend’s boss.

“Our ability to get out and look after a strike was extraordinarily limited — it was an imperfect system,” General Votel said in a telephone interview. “But I believe we always took this seriously and tried to do our best.”

Tips, Intercepts and Strikes

Officially, Talon Anvil never existed. Nearly everything it did was highly classified. The strike cell’s actions in Syria were gleaned from descriptions of top secret reports and interviews with current and former military personnel who interacted with the group and who discussed it on the condition that they not be named.

The strike cell was run by a classified Special Operations unit called Task Force 9 that oversaw the ground offensive in Syria. The task force had multiple missions. Army Green Berets trained allied Syrian Kurdish and Arab forces. Small groups of Delta Force operators embedded with ground forces, and an assault team of Delta commandos were on call to launch ground raids on high-value targets, including the Islamic State leader, Abu Bakr al-Baghdadi.

Most of the firepower, though, was run by Talon Anvil. It worked out of bland office spaces, first in Erbil, Iraq, and then, as the war progressed, in Syria, at a shuttered cement plant in the north, and at a housing complex near the Iraqi border called Green Village, former task force members said.

The cell used tips from allied ground forces, secret electronic intercepts, drone cameras and other information to find enemy targets, then hit them with munitions from drones or called in strikes from other coalition aircraft. It also coordinated air support for allied Kurdish and Arab forces fighting on the ground.

Outwardly, the operators showed few signs that they were military, said a former task force member who worked with the strike cell during the height of the war in 2017. They used first names and no rank or uniforms, and many had bushy beards and went to work in shorts and footwear that included Crocs and Birkenstocks. But from their strike room, they controlled a fleet of Predator and Reaper drones that bristled with precision Hellfire missiles and laser-guided bombs.

The task force had a second strike cell that worked with the C.I.A. to hunt high-value Islamic State leaders. It used similar tools, but often tracked a target for days or weeks, and accounted for a fraction of the strikes.

Both cells were created in 2014 when the Islamic State had overrun large parts of Iraq and Syria. Within a few years, the self-declared caliphate was attacking allies in the Middle East and launching terrorist attacks in Europe. The United States was desperate for a force that could identify enemy targets, and put Delta Force in charge.

Early in the American-led offensive, which was known as Operation Inherent Resolve, the military struggled to function at “the speed of war,” as only high-ranking generals from outside Delta could approve strikes, according to a RAND Corporation report on the air war. Seventy-four percent of sorties returned without dropping any weapons, and the offensive began to stall.

Tactics changed late in 2016 when General Townsend took command and, in an attempt to keep pace with a rapidly expanding offensive, moved the authority to approve strikes down to the level of on-scene commanders.

Within Task Force 9, that authority was effectively pushed even lower, a senior official with extensive experience in Iraq and Syria said, to the senior enlisted Delta operator on shift in the strike room — usually a sergeant first class or master sergeant.

Under the new rules, the strike cell was still required to follow a process of intelligence gathering and risk mitigation to limit harm to civilians before launching a strike. That often meant flying drones over targets for hours to make sure the cell could positively identify enemies and determine whether civilians were in the area.

But the Delta operators were under enormous pressure to protect allied ground troops and move the offensive forward, the former task force member said, and felt hobbled by the safeguards. So in early 2017, they found a way to strike more quickly: self-defense.

Most of Operation Inherent Resolve’s restrictions applied only to offensive strikes. There were far fewer restrictions for defensive strikes that were meant to protect allied forces under imminent threat of harm. So Talon Anvil began claiming that nearly every strike was in self-defense, which enabled them to move quickly with little second-guessing or oversight, even if their targets were miles from any fighting, two former task force members said.

The classified rules of engagement warned that self-defense strikes should not be used to circumvent the more restrictive rules for offensive strikes, two officers with knowledge of the rules said. But for Talon Anvil, there was a tenuous logic to the tactic, one of the former task force members said. If defense rules allowed Talon Anvil to attack an enemy target on the front lines, then why not the same type of target 10 or even 100 miles away that might one day be on the front lines? Soon Talon Anvil was justifying nearly every strike as defensive.

“It’s more expedient to resort to self-defense,” said Mr. Lewis, the former Pentagon adviser. “It’s easier to get approved.”

But speeding up strikes meant less time to gather intelligence and sort enemy fighters from civilians, and the four former military personnel who worked with Talon Anvil said that too often the cell relied on flimsy intelligence from Kurdish and Arab ground forces or rushed to attack with little regard to who might be nearby.

One former task force member said the vast majority of Talon Anvil’s strikes killed only enemy fighters, but that the Delta operators in the strike cell were biased toward hitting and often decided something was an enemy target when there was scant supporting evidence. Part of the problem, he said, was that operators, who rotated through roughly every four months, were trained as elite commandos but had little experience running a strike cell. In addition, he said, the daily demands of overseeing strike after strike seemed to erode operators’ perspective and fray their humanity.

The former Air Force intelligence officer said he saw so many civilian deaths as a result of Talon Anvil’s tactics citing self-defense that he eventually grew jaded and accepted them as part of the job. Even still, some attacks stood out.

In one, he said, Talon Anvil followed three men, all with canvas bags, working in an olive grove near the city of Manbij in the fall of 2016. The men had no weapons, and were not near any fighting, but the strike cell insisted they must be enemy fighters and killed them with a missile.

In another, as civilians were trying to flee fighting in the city of Raqqa in June 2017, scores of people boarded makeshift ferries to cross the Euphrates River. He said the task force claimed the ferries were carrying enemy fighters, and he watched on high-definition video as it hit multiple boats, killing at least 30 civilians, whose bodies drifted away in the green water.

A senior military official with direct knowledge of the task force said that what counted as an “imminent threat” was extremely subjective and Talon Anvil’s senior Delta operators were given broad authority to launch defensive strikes. At times, the official acknowledged, that led to bad strikes, and those who showed poor judgment were removed. But the official emphasized these instances were rare.

Fighters, or Children?

As airstrikes escalated in 2017, a broad array of U.S. partners working with the strike cell grew troubled by its tactics.

The C.I.A. had officers embedded in Task Force 9 to supply intelligence on Islamic State leaders and coordinate strikes. The agency was pursuing high-value individuals, and often tracked them for days using multiple drones, waiting to strike when civilian deaths could be minimized.

The task force did not always like to wait, two former C.I.A. officers said. C.I.A. personnel were shocked when they repeatedly saw the group strike with little regard for civilians. Officers reported their concerns to the Department of Defense’s Inspector General, and the agency’s leadership discussed the issue with top officers at the Joint Special Operations Command, one former C.I.A. officer said.

The officer said he never saw evidence that these concerns were taken seriously.

A C.I.A. spokesman declined to comment.

Talon Anvil also clashed at times with the Air Force intelligence teams based in the United States that helped to analyze the torrent of footage from drones. The Delta operators would push analysts to say they saw evidence such as weapons that could legally justify a strike, even when there was none, the former Air Force intelligence officer said. If one analyst did not see what Delta wanted, Delta would ask for a different one.

Delta Force and analysts sometimes argued over whether figures in the sights of a drone were fighters or children, one of the former task force members said.

All of the footage from the strikes is stored by the military. In an apparent attempt to blunt criticism and undercut potential investigations, Talon Anvil started directing drone cameras away from targets shortly before a strike hit, preventing the collection of video evidence, the former Air Force intelligence officer and one of the former task force members said.

Another Air Force officer, who reviewed dozens of task force strikes where civilians were reportedly killed, said that drone crews were trained to keep cameras on targets so the military could assess damage. Yet he frequently saw cameras jerk away at key moments, as if hit by a wind gust. It was only after seeing the pattern over and over, he said, that he began to believe it was done on purpose.

A Hunt for Targets

One morning before dawn in early March 2017, Talon Anvil sent a Predator drone over a Syrian farming town called Karama to cripple enemy positions in the area in preparation for an offensive by allies a week later.

For the former Air Force intelligence officer, the mission stands out as an example of Talon Anvil’s flawed way of operating, and how military leaders seemed to look the other way.

At about 4 a.m., he said, the drone arrived over the town’s flat-roofed houses. His Air Force intelligence team was watching from a secure operations center in the United States. A Talon Anvil operator typed a message into the chat room the cell shared with intelligence analysts: All civilians have fled the area. Anyone left is an enemy fighter. Find lots of targets for us today because we want to go Winchester.

Going Winchester meant expending all of the drone’s missiles and 500-pound bombs.

As the drone circled, the town appeared to be asleep, the former officer said. Even with infrared sensors, the team did not see movement. Talon Anvil focused in on a building and typed in the chat that a tip from ground forces indicated that the building was an enemy training center. Sensors suggested an enemy cellphone or radio might be in the neighborhood but was unable to pinpoint it to a single block, let alone a single building.

Talon Anvil did not wait for confirmation, and ordered a self-defense strike, the former officer said. The Predator dropped a 500-pound bomb through the roof.

As the smoke cleared, the former officer said, his team stared at their screens in dismay. The infrared cameras showed women and children staggering out of the partly collapsed building, some missing limbs, some dragging the dead.

The intelligence analysts began taking screen shots and tallying the casualties. They sent an initial battle damage assessment to Talon Anvil: 23 dead or severely wounded, 30 lightly wounded, very likely civilians. Talon Anvil paused only long enough to acknowledge the message, the former officer said, then pressed on to the next target.

The former Air Force officer said he immediately reported the civilian casualties to Operation Inherent Resolve’s operations center, then called the center’s liaison officer on the red line. He said he never heard back and saw no evidence that any action was ever taken.

Operation Inherent Resolve made a commitment to investigate and report every case of civilian casualties publicly, but nothing in its reports matches the incident. The true toll of the strike in Karama remains uncertain.

During a five-day window in early March, Operation Inherent Resolve acknowledged that it launched 47 strikes in the region. Satellite images from the time show extensive damage to at least a dozen buildings, including the building that the former officer said he saw bombed. Local media reported that airstrikes in Karama on March 8 and 9 killed between seven and 14 people and wounded 18.

For two years after the strikes, Operation Inherent Resolve said it could not confirm any civilian casualties in the town. Then, in 2019, it acknowledged that one man had been wounded when the coalition struck an enemy fighting position. It gave coordinates a block from the building the former Air Force intelligence officer said he saw destroyed.

In response to questions from The Times this month, a Special Operations official acknowledged its strike cell had hit targets in the town on March 8 and killed 16 fighters, but denied that any civilians had died.

No outside group has ever investigated the secret strike, and it is unclear what steps the military took to determine what happened. The former officer said no military investigators ever contacted him.

The evidence from the strike — the chat room records, bombing coordinates and video — is stored on government servers, the former officer said. But because of the secrecy surrounding Talon Anvil, all of it is classified.

Azmat Khan contributed reporting. Additional production by Christoph Koettl and Drew Jordan.